Spaces:
Build error
Build error
Michael Hu
commited on
Commit
Β·
1f9c751
1
Parent(s):
e3cb97b
Migrate existing TTS providers to infrastructure layer
Browse files- src/domain/models/speech_synthesis_request.py +9 -9
- src/infrastructure/tts/__init__.py +28 -0
- src/infrastructure/tts/cosyvoice2_provider.py +159 -0
- src/infrastructure/tts/dia_provider.py +170 -0
- src/infrastructure/tts/dummy_provider.py +139 -0
- src/infrastructure/tts/kokoro_provider.py +131 -0
- src/infrastructure/tts/provider_factory.py +204 -0
- test_tts_migration.py +137 -0
src/domain/models/speech_synthesis_request.py
CHANGED
@@ -10,7 +10,7 @@ from .voice_settings import VoiceSettings
|
|
10 |
class SpeechSynthesisRequest:
|
11 |
"""Value object representing a speech synthesis request."""
|
12 |
|
13 |
-
|
14 |
voice_settings: VoiceSettings
|
15 |
output_format: str = 'wav'
|
16 |
sample_rate: Optional[int] = None
|
@@ -21,7 +21,7 @@ class SpeechSynthesisRequest:
|
|
21 |
|
22 |
def _validate(self):
|
23 |
"""Validate speech synthesis request properties."""
|
24 |
-
if not isinstance(self.
|
25 |
raise TypeError("Text must be a TextContent instance")
|
26 |
|
27 |
if not isinstance(self.voice_settings, VoiceSettings):
|
@@ -44,8 +44,8 @@ class SpeechSynthesisRequest:
|
|
44 |
raise ValueError("Sample rate must be between 8000 and 192000 Hz")
|
45 |
|
46 |
# Validate that text and voice settings have compatible languages
|
47 |
-
if self.
|
48 |
-
raise ValueError(f"Text language ({self.
|
49 |
|
50 |
@property
|
51 |
def estimated_duration_seconds(self) -> float:
|
@@ -53,12 +53,12 @@ class SpeechSynthesisRequest:
|
|
53 |
# Rough estimation: average speaking rate is about 150-200 words per minute
|
54 |
# Adjusted by speed setting
|
55 |
words_per_minute = 175 / self.voice_settings.speed
|
56 |
-
return (self.
|
57 |
|
58 |
@property
|
59 |
def is_long_text(self) -> bool:
|
60 |
"""Check if the text is considered long for TTS processing."""
|
61 |
-
return self.
|
62 |
|
63 |
@property
|
64 |
def effective_sample_rate(self) -> int:
|
@@ -68,7 +68,7 @@ class SpeechSynthesisRequest:
|
|
68 |
def with_output_format(self, output_format: str) -> 'SpeechSynthesisRequest':
|
69 |
"""Create a new SpeechSynthesisRequest with different output format."""
|
70 |
return SpeechSynthesisRequest(
|
71 |
-
|
72 |
voice_settings=self.voice_settings,
|
73 |
output_format=output_format,
|
74 |
sample_rate=self.sample_rate
|
@@ -77,7 +77,7 @@ class SpeechSynthesisRequest:
|
|
77 |
def with_sample_rate(self, sample_rate: Optional[int]) -> 'SpeechSynthesisRequest':
|
78 |
"""Create a new SpeechSynthesisRequest with different sample rate."""
|
79 |
return SpeechSynthesisRequest(
|
80 |
-
|
81 |
voice_settings=self.voice_settings,
|
82 |
output_format=self.output_format,
|
83 |
sample_rate=sample_rate
|
@@ -86,7 +86,7 @@ class SpeechSynthesisRequest:
|
|
86 |
def with_voice_settings(self, voice_settings: VoiceSettings) -> 'SpeechSynthesisRequest':
|
87 |
"""Create a new SpeechSynthesisRequest with different voice settings."""
|
88 |
return SpeechSynthesisRequest(
|
89 |
-
|
90 |
voice_settings=voice_settings,
|
91 |
output_format=self.output_format,
|
92 |
sample_rate=self.sample_rate
|
|
|
10 |
class SpeechSynthesisRequest:
|
11 |
"""Value object representing a speech synthesis request."""
|
12 |
|
13 |
+
text_content: TextContent
|
14 |
voice_settings: VoiceSettings
|
15 |
output_format: str = 'wav'
|
16 |
sample_rate: Optional[int] = None
|
|
|
21 |
|
22 |
def _validate(self):
|
23 |
"""Validate speech synthesis request properties."""
|
24 |
+
if not isinstance(self.text_content, TextContent):
|
25 |
raise TypeError("Text must be a TextContent instance")
|
26 |
|
27 |
if not isinstance(self.voice_settings, VoiceSettings):
|
|
|
44 |
raise ValueError("Sample rate must be between 8000 and 192000 Hz")
|
45 |
|
46 |
# Validate that text and voice settings have compatible languages
|
47 |
+
if self.text_content.language != self.voice_settings.language:
|
48 |
+
raise ValueError(f"Text language ({self.text_content.language}) must match voice language ({self.voice_settings.language})")
|
49 |
|
50 |
@property
|
51 |
def estimated_duration_seconds(self) -> float:
|
|
|
53 |
# Rough estimation: average speaking rate is about 150-200 words per minute
|
54 |
# Adjusted by speed setting
|
55 |
words_per_minute = 175 / self.voice_settings.speed
|
56 |
+
return (self.text_content.word_count / words_per_minute) * 60
|
57 |
|
58 |
@property
|
59 |
def is_long_text(self) -> bool:
|
60 |
"""Check if the text is considered long for TTS processing."""
|
61 |
+
return self.text_content.character_count > 5000
|
62 |
|
63 |
@property
|
64 |
def effective_sample_rate(self) -> int:
|
|
|
68 |
def with_output_format(self, output_format: str) -> 'SpeechSynthesisRequest':
|
69 |
"""Create a new SpeechSynthesisRequest with different output format."""
|
70 |
return SpeechSynthesisRequest(
|
71 |
+
text_content=self.text_content,
|
72 |
voice_settings=self.voice_settings,
|
73 |
output_format=output_format,
|
74 |
sample_rate=self.sample_rate
|
|
|
77 |
def with_sample_rate(self, sample_rate: Optional[int]) -> 'SpeechSynthesisRequest':
|
78 |
"""Create a new SpeechSynthesisRequest with different sample rate."""
|
79 |
return SpeechSynthesisRequest(
|
80 |
+
text_content=self.text_content,
|
81 |
voice_settings=self.voice_settings,
|
82 |
output_format=self.output_format,
|
83 |
sample_rate=sample_rate
|
|
|
86 |
def with_voice_settings(self, voice_settings: VoiceSettings) -> 'SpeechSynthesisRequest':
|
87 |
"""Create a new SpeechSynthesisRequest with different voice settings."""
|
88 |
return SpeechSynthesisRequest(
|
89 |
+
text_content=self.text_content,
|
90 |
voice_settings=voice_settings,
|
91 |
output_format=self.output_format,
|
92 |
sample_rate=self.sample_rate
|
src/infrastructure/tts/__init__.py
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""TTS provider implementations."""
|
2 |
+
|
3 |
+
from .provider_factory import TTSProviderFactory
|
4 |
+
from .dummy_provider import DummyTTSProvider
|
5 |
+
|
6 |
+
# Try to import optional providers
|
7 |
+
try:
|
8 |
+
from .kokoro_provider import KokoroTTSProvider
|
9 |
+
except ImportError:
|
10 |
+
KokoroTTSProvider = None
|
11 |
+
|
12 |
+
try:
|
13 |
+
from .dia_provider import DiaTTSProvider
|
14 |
+
except ImportError:
|
15 |
+
DiaTTSProvider = None
|
16 |
+
|
17 |
+
try:
|
18 |
+
from .cosyvoice2_provider import CosyVoice2TTSProvider
|
19 |
+
except ImportError:
|
20 |
+
CosyVoice2TTSProvider = None
|
21 |
+
|
22 |
+
__all__ = [
|
23 |
+
'TTSProviderFactory',
|
24 |
+
'DummyTTSProvider',
|
25 |
+
'KokoroTTSProvider',
|
26 |
+
'DiaTTSProvider',
|
27 |
+
'CosyVoice2TTSProvider'
|
28 |
+
]
|
src/infrastructure/tts/cosyvoice2_provider.py
ADDED
@@ -0,0 +1,159 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""CosyVoice2 TTS provider implementation."""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
import numpy as np
|
5 |
+
import soundfile as sf
|
6 |
+
import io
|
7 |
+
from typing import Iterator, TYPE_CHECKING
|
8 |
+
|
9 |
+
if TYPE_CHECKING:
|
10 |
+
from ...domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
11 |
+
|
12 |
+
from ..base.tts_provider_base import TTSProviderBase
|
13 |
+
from ...domain.exceptions import SpeechSynthesisException
|
14 |
+
|
15 |
+
logger = logging.getLogger(__name__)
|
16 |
+
|
17 |
+
# Flag to track CosyVoice2 availability
|
18 |
+
COSYVOICE2_AVAILABLE = False
|
19 |
+
DEFAULT_SAMPLE_RATE = 24000
|
20 |
+
|
21 |
+
# Try to import CosyVoice2 dependencies
|
22 |
+
try:
|
23 |
+
import torch
|
24 |
+
# Import CosyVoice2 - assuming it's installed and has a similar API to Dia
|
25 |
+
# since they're both from nari-labs according to the GitHub link
|
26 |
+
from cosyvoice2.model import CosyVoice2
|
27 |
+
COSYVOICE2_AVAILABLE = True
|
28 |
+
logger.info("CosyVoice2 TTS engine is available")
|
29 |
+
except ImportError:
|
30 |
+
logger.warning("CosyVoice2 TTS engine is not available")
|
31 |
+
except ModuleNotFoundError as e:
|
32 |
+
logger.warning(f"CosyVoice2 TTS engine is not available: {str(e)}")
|
33 |
+
COSYVOICE2_AVAILABLE = False
|
34 |
+
|
35 |
+
|
36 |
+
class CosyVoice2TTSProvider(TTSProviderBase):
|
37 |
+
"""CosyVoice2 TTS provider implementation."""
|
38 |
+
|
39 |
+
def __init__(self, lang_code: str = 'z'):
|
40 |
+
"""Initialize the CosyVoice2 TTS provider."""
|
41 |
+
super().__init__(
|
42 |
+
provider_name="CosyVoice2",
|
43 |
+
supported_languages=['en', 'z'] # CosyVoice2 supports English and multilingual
|
44 |
+
)
|
45 |
+
self.lang_code = lang_code
|
46 |
+
self.model = None
|
47 |
+
|
48 |
+
def _ensure_model(self):
|
49 |
+
"""Ensure the model is loaded."""
|
50 |
+
if self.model is None and COSYVOICE2_AVAILABLE:
|
51 |
+
try:
|
52 |
+
import torch
|
53 |
+
from cosyvoice2.model import CosyVoice2
|
54 |
+
self.model = CosyVoice2.from_pretrained()
|
55 |
+
logger.info("CosyVoice2 model successfully loaded")
|
56 |
+
except ImportError as e:
|
57 |
+
logger.error(f"Failed to import CosyVoice2 dependencies: {str(e)}")
|
58 |
+
self.model = None
|
59 |
+
except FileNotFoundError as e:
|
60 |
+
logger.error(f"Failed to load CosyVoice2 model files: {str(e)}")
|
61 |
+
self.model = None
|
62 |
+
except Exception as e:
|
63 |
+
logger.error(f"Failed to initialize CosyVoice2 model: {str(e)}")
|
64 |
+
self.model = None
|
65 |
+
return self.model is not None
|
66 |
+
|
67 |
+
def is_available(self) -> bool:
|
68 |
+
"""Check if CosyVoice2 TTS is available."""
|
69 |
+
return COSYVOICE2_AVAILABLE and self._ensure_model()
|
70 |
+
|
71 |
+
def get_available_voices(self) -> list[str]:
|
72 |
+
"""Get available voices for CosyVoice2."""
|
73 |
+
# CosyVoice2 typically uses a default voice
|
74 |
+
return ['default']
|
75 |
+
|
76 |
+
def _generate_audio(self, request: 'SpeechSynthesisRequest') -> tuple[bytes, int]:
|
77 |
+
"""Generate audio using CosyVoice2 TTS."""
|
78 |
+
if not self.is_available():
|
79 |
+
raise SpeechSynthesisException("CosyVoice2 TTS engine is not available")
|
80 |
+
|
81 |
+
try:
|
82 |
+
import torch
|
83 |
+
|
84 |
+
# Extract parameters from request
|
85 |
+
text = request.text_content.text
|
86 |
+
|
87 |
+
# Generate audio using CosyVoice2
|
88 |
+
with torch.inference_mode():
|
89 |
+
# Assuming CosyVoice2 has a similar API to Dia
|
90 |
+
output_audio_np = self.model.generate(
|
91 |
+
text,
|
92 |
+
max_tokens=None,
|
93 |
+
cfg_scale=3.0,
|
94 |
+
temperature=1.3,
|
95 |
+
top_p=0.95,
|
96 |
+
use_torch_compile=False,
|
97 |
+
verbose=False
|
98 |
+
)
|
99 |
+
|
100 |
+
if output_audio_np is None:
|
101 |
+
raise SpeechSynthesisException("CosyVoice2 model returned None for audio output")
|
102 |
+
|
103 |
+
# Convert numpy array to bytes
|
104 |
+
audio_bytes = self._numpy_to_bytes(output_audio_np, sample_rate=DEFAULT_SAMPLE_RATE)
|
105 |
+
return audio_bytes, DEFAULT_SAMPLE_RATE
|
106 |
+
|
107 |
+
except Exception as e:
|
108 |
+
self._handle_provider_error(e, "audio generation")
|
109 |
+
|
110 |
+
def _generate_audio_stream(self, request: 'SpeechSynthesisRequest') -> Iterator[tuple[bytes, int, bool]]:
|
111 |
+
"""Generate audio stream using CosyVoice2 TTS."""
|
112 |
+
if not self.is_available():
|
113 |
+
raise SpeechSynthesisException("CosyVoice2 TTS engine is not available")
|
114 |
+
|
115 |
+
try:
|
116 |
+
import torch
|
117 |
+
|
118 |
+
# Extract parameters from request
|
119 |
+
text = request.text_content.text
|
120 |
+
|
121 |
+
# Generate audio using CosyVoice2
|
122 |
+
with torch.inference_mode():
|
123 |
+
# Assuming CosyVoice2 has a similar API to Dia
|
124 |
+
output_audio_np = self.model.generate(
|
125 |
+
text,
|
126 |
+
max_tokens=None,
|
127 |
+
cfg_scale=3.0,
|
128 |
+
temperature=1.3,
|
129 |
+
top_p=0.95,
|
130 |
+
use_torch_compile=False,
|
131 |
+
verbose=False
|
132 |
+
)
|
133 |
+
|
134 |
+
if output_audio_np is None:
|
135 |
+
raise SpeechSynthesisException("CosyVoice2 model returned None for audio output")
|
136 |
+
|
137 |
+
# Convert numpy array to bytes
|
138 |
+
audio_bytes = self._numpy_to_bytes(output_audio_np, sample_rate=DEFAULT_SAMPLE_RATE)
|
139 |
+
# CosyVoice2 generates complete audio in one go
|
140 |
+
yield audio_bytes, DEFAULT_SAMPLE_RATE, True
|
141 |
+
|
142 |
+
except Exception as e:
|
143 |
+
self._handle_provider_error(e, "streaming audio generation")
|
144 |
+
|
145 |
+
def _numpy_to_bytes(self, audio_array: np.ndarray, sample_rate: int) -> bytes:
|
146 |
+
"""Convert numpy audio array to bytes."""
|
147 |
+
try:
|
148 |
+
# Create an in-memory buffer
|
149 |
+
buffer = io.BytesIO()
|
150 |
+
|
151 |
+
# Write audio data to buffer as WAV
|
152 |
+
sf.write(buffer, audio_array, sample_rate, format='WAV')
|
153 |
+
|
154 |
+
# Get bytes from buffer
|
155 |
+
buffer.seek(0)
|
156 |
+
return buffer.read()
|
157 |
+
|
158 |
+
except Exception as e:
|
159 |
+
raise SpeechSynthesisException(f"Failed to convert audio to bytes: {str(e)}") from e
|
src/infrastructure/tts/dia_provider.py
ADDED
@@ -0,0 +1,170 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Dia TTS provider implementation."""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
import numpy as np
|
5 |
+
import soundfile as sf
|
6 |
+
import io
|
7 |
+
from typing import Iterator, TYPE_CHECKING
|
8 |
+
|
9 |
+
if TYPE_CHECKING:
|
10 |
+
from ...domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
11 |
+
|
12 |
+
from ..base.tts_provider_base import TTSProviderBase
|
13 |
+
from ...domain.exceptions import SpeechSynthesisException
|
14 |
+
|
15 |
+
logger = logging.getLogger(__name__)
|
16 |
+
|
17 |
+
# Flag to track Dia availability
|
18 |
+
DIA_AVAILABLE = False
|
19 |
+
DEFAULT_SAMPLE_RATE = 24000
|
20 |
+
|
21 |
+
# Try to import Dia dependencies
|
22 |
+
try:
|
23 |
+
import torch
|
24 |
+
from dia.model import Dia
|
25 |
+
DIA_AVAILABLE = True
|
26 |
+
logger.info("Dia TTS engine is available")
|
27 |
+
except ImportError:
|
28 |
+
logger.warning("Dia TTS engine is not available")
|
29 |
+
except ModuleNotFoundError as e:
|
30 |
+
if "dac" in str(e):
|
31 |
+
logger.warning("Dia TTS engine is not available due to missing 'dac' module")
|
32 |
+
else:
|
33 |
+
logger.warning(f"Dia TTS engine is not available: {str(e)}")
|
34 |
+
DIA_AVAILABLE = False
|
35 |
+
|
36 |
+
|
37 |
+
class DiaTTSProvider(TTSProviderBase):
|
38 |
+
"""Dia TTS provider implementation."""
|
39 |
+
|
40 |
+
def __init__(self, lang_code: str = 'z'):
|
41 |
+
"""Initialize the Dia TTS provider."""
|
42 |
+
super().__init__(
|
43 |
+
provider_name="Dia",
|
44 |
+
supported_languages=['en', 'z'] # Dia supports English and multilingual
|
45 |
+
)
|
46 |
+
self.lang_code = lang_code
|
47 |
+
self.model = None
|
48 |
+
|
49 |
+
def _ensure_model(self):
|
50 |
+
"""Ensure the model is loaded."""
|
51 |
+
if self.model is None and DIA_AVAILABLE:
|
52 |
+
try:
|
53 |
+
import torch
|
54 |
+
from dia.model import Dia
|
55 |
+
self.model = Dia.from_pretrained()
|
56 |
+
logger.info("Dia model successfully loaded")
|
57 |
+
except ImportError as e:
|
58 |
+
logger.error(f"Failed to import Dia dependencies: {str(e)}")
|
59 |
+
self.model = None
|
60 |
+
except FileNotFoundError as e:
|
61 |
+
logger.error(f"Failed to load Dia model files: {str(e)}")
|
62 |
+
self.model = None
|
63 |
+
except Exception as e:
|
64 |
+
logger.error(f"Failed to initialize Dia model: {str(e)}")
|
65 |
+
self.model = None
|
66 |
+
return self.model is not None
|
67 |
+
|
68 |
+
def is_available(self) -> bool:
|
69 |
+
"""Check if Dia TTS is available."""
|
70 |
+
return DIA_AVAILABLE and self._ensure_model()
|
71 |
+
|
72 |
+
def get_available_voices(self) -> list[str]:
|
73 |
+
"""Get available voices for Dia."""
|
74 |
+
# Dia typically uses a default voice
|
75 |
+
return ['default']
|
76 |
+
|
77 |
+
def _generate_audio(self, request: 'SpeechSynthesisRequest') -> tuple[bytes, int]:
|
78 |
+
"""Generate audio using Dia TTS."""
|
79 |
+
if not self.is_available():
|
80 |
+
raise SpeechSynthesisException("Dia TTS engine is not available")
|
81 |
+
|
82 |
+
try:
|
83 |
+
import torch
|
84 |
+
|
85 |
+
# Extract parameters from request
|
86 |
+
text = request.text_content.text
|
87 |
+
|
88 |
+
# Generate audio using Dia
|
89 |
+
with torch.inference_mode():
|
90 |
+
output_audio_np = self.model.generate(
|
91 |
+
text,
|
92 |
+
max_tokens=None,
|
93 |
+
cfg_scale=3.0,
|
94 |
+
temperature=1.3,
|
95 |
+
top_p=0.95,
|
96 |
+
cfg_filter_top_k=35,
|
97 |
+
use_torch_compile=False,
|
98 |
+
verbose=False
|
99 |
+
)
|
100 |
+
|
101 |
+
if output_audio_np is None:
|
102 |
+
raise SpeechSynthesisException("Dia model returned None for audio output")
|
103 |
+
|
104 |
+
# Convert numpy array to bytes
|
105 |
+
audio_bytes = self._numpy_to_bytes(output_audio_np, sample_rate=DEFAULT_SAMPLE_RATE)
|
106 |
+
return audio_bytes, DEFAULT_SAMPLE_RATE
|
107 |
+
|
108 |
+
except ModuleNotFoundError as e:
|
109 |
+
if "dac" in str(e):
|
110 |
+
raise SpeechSynthesisException("Dia TTS engine failed due to missing 'dac' module") from e
|
111 |
+
else:
|
112 |
+
self._handle_provider_error(e, "audio generation")
|
113 |
+
except Exception as e:
|
114 |
+
self._handle_provider_error(e, "audio generation")
|
115 |
+
|
116 |
+
def _generate_audio_stream(self, request: 'SpeechSynthesisRequest') -> Iterator[tuple[bytes, int, bool]]:
|
117 |
+
"""Generate audio stream using Dia TTS."""
|
118 |
+
if not self.is_available():
|
119 |
+
raise SpeechSynthesisException("Dia TTS engine is not available")
|
120 |
+
|
121 |
+
try:
|
122 |
+
import torch
|
123 |
+
|
124 |
+
# Extract parameters from request
|
125 |
+
text = request.text_content.text
|
126 |
+
|
127 |
+
# Generate audio using Dia
|
128 |
+
with torch.inference_mode():
|
129 |
+
output_audio_np = self.model.generate(
|
130 |
+
text,
|
131 |
+
max_tokens=None,
|
132 |
+
cfg_scale=3.0,
|
133 |
+
temperature=1.3,
|
134 |
+
top_p=0.95,
|
135 |
+
cfg_filter_top_k=35,
|
136 |
+
use_torch_compile=False,
|
137 |
+
verbose=False
|
138 |
+
)
|
139 |
+
|
140 |
+
if output_audio_np is None:
|
141 |
+
raise SpeechSynthesisException("Dia model returned None for audio output")
|
142 |
+
|
143 |
+
# Convert numpy array to bytes
|
144 |
+
audio_bytes = self._numpy_to_bytes(output_audio_np, sample_rate=DEFAULT_SAMPLE_RATE)
|
145 |
+
# Dia generates complete audio in one go
|
146 |
+
yield audio_bytes, DEFAULT_SAMPLE_RATE, True
|
147 |
+
|
148 |
+
except ModuleNotFoundError as e:
|
149 |
+
if "dac" in str(e):
|
150 |
+
raise SpeechSynthesisException("Dia TTS engine failed due to missing 'dac' module") from e
|
151 |
+
else:
|
152 |
+
self._handle_provider_error(e, "streaming audio generation")
|
153 |
+
except Exception as e:
|
154 |
+
self._handle_provider_error(e, "streaming audio generation")
|
155 |
+
|
156 |
+
def _numpy_to_bytes(self, audio_array: np.ndarray, sample_rate: int) -> bytes:
|
157 |
+
"""Convert numpy audio array to bytes."""
|
158 |
+
try:
|
159 |
+
# Create an in-memory buffer
|
160 |
+
buffer = io.BytesIO()
|
161 |
+
|
162 |
+
# Write audio data to buffer as WAV
|
163 |
+
sf.write(buffer, audio_array, sample_rate, format='WAV')
|
164 |
+
|
165 |
+
# Get bytes from buffer
|
166 |
+
buffer.seek(0)
|
167 |
+
return buffer.read()
|
168 |
+
|
169 |
+
except Exception as e:
|
170 |
+
raise SpeechSynthesisException(f"Failed to convert audio to bytes: {str(e)}") from e
|
src/infrastructure/tts/dummy_provider.py
ADDED
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Dummy TTS provider implementation for testing and fallback."""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
import numpy as np
|
5 |
+
import soundfile as sf
|
6 |
+
import io
|
7 |
+
from typing import Iterator, TYPE_CHECKING
|
8 |
+
|
9 |
+
if TYPE_CHECKING:
|
10 |
+
from ...domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
11 |
+
|
12 |
+
from ..base.tts_provider_base import TTSProviderBase
|
13 |
+
from ...domain.exceptions import SpeechSynthesisException
|
14 |
+
|
15 |
+
logger = logging.getLogger(__name__)
|
16 |
+
|
17 |
+
|
18 |
+
class DummyTTSProvider(TTSProviderBase):
|
19 |
+
"""Dummy TTS provider that generates sine wave audio for testing."""
|
20 |
+
|
21 |
+
def __init__(self):
|
22 |
+
"""Initialize the Dummy TTS provider."""
|
23 |
+
super().__init__(
|
24 |
+
provider_name="Dummy",
|
25 |
+
supported_languages=['en', 'es', 'fr', 'de', 'it', 'pt', 'ru', 'ja', 'ko', 'zh']
|
26 |
+
)
|
27 |
+
|
28 |
+
def is_available(self) -> bool:
|
29 |
+
"""Dummy TTS is always available."""
|
30 |
+
return True
|
31 |
+
|
32 |
+
def get_available_voices(self) -> list[str]:
|
33 |
+
"""Get available voices for Dummy TTS."""
|
34 |
+
return ['default', 'male', 'female', 'robot']
|
35 |
+
|
36 |
+
def _generate_audio(self, request: 'SpeechSynthesisRequest') -> tuple[bytes, int]:
|
37 |
+
"""Generate dummy sine wave audio."""
|
38 |
+
try:
|
39 |
+
# Extract parameters from request
|
40 |
+
text = request.text_content.text
|
41 |
+
speed = request.voice_settings.speed
|
42 |
+
|
43 |
+
# Generate a simple sine wave based on text length and speed
|
44 |
+
sample_rate = 24000
|
45 |
+
# Rough approximation of speech duration adjusted by speed
|
46 |
+
duration = min(len(text) / (20 * speed), 10)
|
47 |
+
|
48 |
+
# Create time array
|
49 |
+
t = np.linspace(0, duration, int(sample_rate * duration), endpoint=False)
|
50 |
+
|
51 |
+
# Generate sine wave (440 Hz base frequency)
|
52 |
+
frequency = 440
|
53 |
+
audio = 0.5 * np.sin(2 * np.pi * frequency * t)
|
54 |
+
|
55 |
+
# Add some variation based on voice setting
|
56 |
+
voice = request.voice_settings.voice_id
|
57 |
+
if voice == 'male':
|
58 |
+
# Lower frequency for male voice
|
59 |
+
audio = 0.5 * np.sin(2 * np.pi * 220 * t)
|
60 |
+
elif voice == 'female':
|
61 |
+
# Higher frequency for female voice
|
62 |
+
audio = 0.5 * np.sin(2 * np.pi * 660 * t)
|
63 |
+
elif voice == 'robot':
|
64 |
+
# Square wave for robot voice
|
65 |
+
audio = 0.5 * np.sign(np.sin(2 * np.pi * 440 * t))
|
66 |
+
|
67 |
+
# Convert to bytes
|
68 |
+
audio_bytes = self._numpy_to_bytes(audio, sample_rate)
|
69 |
+
|
70 |
+
logger.info(f"Generated dummy audio: duration={duration:.2f}s, voice={voice}")
|
71 |
+
return audio_bytes, sample_rate
|
72 |
+
|
73 |
+
except Exception as e:
|
74 |
+
self._handle_provider_error(e, "dummy audio generation")
|
75 |
+
|
76 |
+
def _generate_audio_stream(self, request: 'SpeechSynthesisRequest') -> Iterator[tuple[bytes, int, bool]]:
|
77 |
+
"""Generate dummy sine wave audio stream."""
|
78 |
+
try:
|
79 |
+
# Extract parameters from request
|
80 |
+
text = request.text_content.text
|
81 |
+
speed = request.voice_settings.speed
|
82 |
+
|
83 |
+
# Generate audio in chunks
|
84 |
+
sample_rate = 24000
|
85 |
+
chunk_duration = 1.0 # 1 second chunks
|
86 |
+
total_duration = min(len(text) / (20 * speed), 10)
|
87 |
+
|
88 |
+
chunks_count = int(np.ceil(total_duration / chunk_duration))
|
89 |
+
|
90 |
+
for chunk_idx in range(chunks_count):
|
91 |
+
start_time = chunk_idx * chunk_duration
|
92 |
+
end_time = min((chunk_idx + 1) * chunk_duration, total_duration)
|
93 |
+
actual_duration = end_time - start_time
|
94 |
+
|
95 |
+
if actual_duration <= 0:
|
96 |
+
break
|
97 |
+
|
98 |
+
# Create time array for this chunk
|
99 |
+
t = np.linspace(0, actual_duration, int(sample_rate * actual_duration), endpoint=False)
|
100 |
+
|
101 |
+
# Generate sine wave
|
102 |
+
frequency = 440
|
103 |
+
audio = 0.5 * np.sin(2 * np.pi * frequency * t)
|
104 |
+
|
105 |
+
# Apply voice variations
|
106 |
+
voice = request.voice_settings.voice_id
|
107 |
+
if voice == 'male':
|
108 |
+
audio = 0.5 * np.sin(2 * np.pi * 220 * t)
|
109 |
+
elif voice == 'female':
|
110 |
+
audio = 0.5 * np.sin(2 * np.pi * 660 * t)
|
111 |
+
elif voice == 'robot':
|
112 |
+
audio = 0.5 * np.sign(np.sin(2 * np.pi * 440 * t))
|
113 |
+
|
114 |
+
# Convert to bytes
|
115 |
+
audio_bytes = self._numpy_to_bytes(audio, sample_rate)
|
116 |
+
|
117 |
+
# Check if this is the final chunk
|
118 |
+
is_final = (chunk_idx == chunks_count - 1)
|
119 |
+
|
120 |
+
yield audio_bytes, sample_rate, is_final
|
121 |
+
|
122 |
+
except Exception as e:
|
123 |
+
self._handle_provider_error(e, "dummy streaming audio generation")
|
124 |
+
|
125 |
+
def _numpy_to_bytes(self, audio_array: np.ndarray, sample_rate: int) -> bytes:
|
126 |
+
"""Convert numpy audio array to bytes."""
|
127 |
+
try:
|
128 |
+
# Create an in-memory buffer
|
129 |
+
buffer = io.BytesIO()
|
130 |
+
|
131 |
+
# Write audio data to buffer as WAV
|
132 |
+
sf.write(buffer, audio_array, sample_rate, format='WAV')
|
133 |
+
|
134 |
+
# Get bytes from buffer
|
135 |
+
buffer.seek(0)
|
136 |
+
return buffer.read()
|
137 |
+
|
138 |
+
except Exception as e:
|
139 |
+
raise SpeechSynthesisException(f"Failed to convert audio to bytes: {str(e)}") from e
|
src/infrastructure/tts/kokoro_provider.py
ADDED
@@ -0,0 +1,131 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Kokoro TTS provider implementation."""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
import numpy as np
|
5 |
+
import soundfile as sf
|
6 |
+
import io
|
7 |
+
from typing import Iterator, TYPE_CHECKING
|
8 |
+
|
9 |
+
if TYPE_CHECKING:
|
10 |
+
from ...domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
11 |
+
|
12 |
+
from ..base.tts_provider_base import TTSProviderBase
|
13 |
+
from ...domain.exceptions import SpeechSynthesisException
|
14 |
+
|
15 |
+
logger = logging.getLogger(__name__)
|
16 |
+
|
17 |
+
# Flag to track Kokoro availability
|
18 |
+
KOKORO_AVAILABLE = False
|
19 |
+
|
20 |
+
# Try to import Kokoro
|
21 |
+
try:
|
22 |
+
from kokoro import KPipeline
|
23 |
+
KOKORO_AVAILABLE = True
|
24 |
+
logger.info("Kokoro TTS engine is available")
|
25 |
+
except ImportError:
|
26 |
+
logger.warning("Kokoro TTS engine is not available")
|
27 |
+
except Exception as e:
|
28 |
+
logger.error(f"Kokoro import failed with unexpected error: {str(e)}")
|
29 |
+
KOKORO_AVAILABLE = False
|
30 |
+
|
31 |
+
|
32 |
+
class KokoroTTSProvider(TTSProviderBase):
|
33 |
+
"""Kokoro TTS provider implementation."""
|
34 |
+
|
35 |
+
def __init__(self, lang_code: str = 'z'):
|
36 |
+
"""Initialize the Kokoro TTS provider."""
|
37 |
+
super().__init__(
|
38 |
+
provider_name="Kokoro",
|
39 |
+
supported_languages=['en', 'z'] # Kokoro supports English and multilingual
|
40 |
+
)
|
41 |
+
self.lang_code = lang_code
|
42 |
+
self.pipeline = None
|
43 |
+
|
44 |
+
def _ensure_pipeline(self):
|
45 |
+
"""Ensure the pipeline is loaded."""
|
46 |
+
if self.pipeline is None and KOKORO_AVAILABLE:
|
47 |
+
try:
|
48 |
+
self.pipeline = KPipeline(lang_code=self.lang_code)
|
49 |
+
logger.info("Kokoro pipeline successfully loaded")
|
50 |
+
except Exception as e:
|
51 |
+
logger.error(f"Failed to initialize Kokoro pipeline: {str(e)}")
|
52 |
+
self.pipeline = None
|
53 |
+
return self.pipeline is not None
|
54 |
+
|
55 |
+
def is_available(self) -> bool:
|
56 |
+
"""Check if Kokoro TTS is available."""
|
57 |
+
return KOKORO_AVAILABLE and self._ensure_pipeline()
|
58 |
+
|
59 |
+
def get_available_voices(self) -> list[str]:
|
60 |
+
"""Get available voices for Kokoro."""
|
61 |
+
# Common Kokoro voices based on the original implementation
|
62 |
+
return [
|
63 |
+
'af_heart', 'af_bella', 'af_sarah', 'af_nicole',
|
64 |
+
'am_adam', 'am_michael', 'bf_emma', 'bf_isabella'
|
65 |
+
]
|
66 |
+
|
67 |
+
def _generate_audio(self, request: 'SpeechSynthesisRequest') -> tuple[bytes, int]:
|
68 |
+
"""Generate audio using Kokoro TTS."""
|
69 |
+
if not self.is_available():
|
70 |
+
raise SpeechSynthesisException("Kokoro TTS engine is not available")
|
71 |
+
|
72 |
+
try:
|
73 |
+
# Extract parameters from request
|
74 |
+
text = request.text_content.text
|
75 |
+
voice = request.voice_settings.voice_id
|
76 |
+
speed = request.voice_settings.speed
|
77 |
+
|
78 |
+
# Generate speech using Kokoro
|
79 |
+
generator = self.pipeline(text, voice=voice, speed=speed)
|
80 |
+
|
81 |
+
for _, _, audio in generator:
|
82 |
+
# Convert numpy array to bytes
|
83 |
+
audio_bytes = self._numpy_to_bytes(audio, sample_rate=24000)
|
84 |
+
return audio_bytes, 24000
|
85 |
+
|
86 |
+
raise SpeechSynthesisException("Kokoro failed to generate audio")
|
87 |
+
|
88 |
+
except Exception as e:
|
89 |
+
self._handle_provider_error(e, "audio generation")
|
90 |
+
|
91 |
+
def _generate_audio_stream(self, request: 'SpeechSynthesisRequest') -> Iterator[tuple[bytes, int, bool]]:
|
92 |
+
"""Generate audio stream using Kokoro TTS."""
|
93 |
+
if not self.is_available():
|
94 |
+
raise SpeechSynthesisException("Kokoro TTS engine is not available")
|
95 |
+
|
96 |
+
try:
|
97 |
+
# Extract parameters from request
|
98 |
+
text = request.text_content.text
|
99 |
+
voice = request.voice_settings.voice_id
|
100 |
+
speed = request.voice_settings.speed
|
101 |
+
|
102 |
+
# Generate speech stream using Kokoro
|
103 |
+
generator = self.pipeline(text, voice=voice, speed=speed)
|
104 |
+
|
105 |
+
chunk_count = 0
|
106 |
+
for _, _, audio in generator:
|
107 |
+
chunk_count += 1
|
108 |
+
# Convert numpy array to bytes
|
109 |
+
audio_bytes = self._numpy_to_bytes(audio, sample_rate=24000)
|
110 |
+
# Assume this is the final chunk for now (Kokoro typically generates one chunk)
|
111 |
+
is_final = True
|
112 |
+
yield audio_bytes, 24000, is_final
|
113 |
+
|
114 |
+
except Exception as e:
|
115 |
+
self._handle_provider_error(e, "streaming audio generation")
|
116 |
+
|
117 |
+
def _numpy_to_bytes(self, audio_array: np.ndarray, sample_rate: int) -> bytes:
|
118 |
+
"""Convert numpy audio array to bytes."""
|
119 |
+
try:
|
120 |
+
# Create an in-memory buffer
|
121 |
+
buffer = io.BytesIO()
|
122 |
+
|
123 |
+
# Write audio data to buffer as WAV
|
124 |
+
sf.write(buffer, audio_array, sample_rate, format='WAV')
|
125 |
+
|
126 |
+
# Get bytes from buffer
|
127 |
+
buffer.seek(0)
|
128 |
+
return buffer.read()
|
129 |
+
|
130 |
+
except Exception as e:
|
131 |
+
raise SpeechSynthesisException(f"Failed to convert audio to bytes: {str(e)}") from e
|
src/infrastructure/tts/provider_factory.py
ADDED
@@ -0,0 +1,204 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""TTS provider factory for creating and managing TTS providers."""
|
2 |
+
|
3 |
+
import logging
|
4 |
+
from typing import Dict, List, Optional, Type
|
5 |
+
from ..base.tts_provider_base import TTSProviderBase
|
6 |
+
from ...domain.exceptions import SpeechSynthesisException
|
7 |
+
|
8 |
+
logger = logging.getLogger(__name__)
|
9 |
+
|
10 |
+
|
11 |
+
class TTSProviderFactory:
|
12 |
+
"""Factory for creating and managing TTS providers."""
|
13 |
+
|
14 |
+
def __init__(self):
|
15 |
+
"""Initialize the TTS provider factory."""
|
16 |
+
self._providers: Dict[str, Type[TTSProviderBase]] = {}
|
17 |
+
self._provider_instances: Dict[str, TTSProviderBase] = {}
|
18 |
+
self._register_default_providers()
|
19 |
+
|
20 |
+
def _register_default_providers(self):
|
21 |
+
"""Register all available TTS providers."""
|
22 |
+
# Import providers dynamically to avoid import errors if dependencies are missing
|
23 |
+
|
24 |
+
# Always register dummy provider as fallback
|
25 |
+
from .dummy_provider import DummyTTSProvider
|
26 |
+
self._providers['dummy'] = DummyTTSProvider
|
27 |
+
|
28 |
+
# Try to register Kokoro provider
|
29 |
+
try:
|
30 |
+
from .kokoro_provider import KokoroTTSProvider
|
31 |
+
self._providers['kokoro'] = KokoroTTSProvider
|
32 |
+
logger.info("Registered Kokoro TTS provider")
|
33 |
+
except ImportError as e:
|
34 |
+
logger.debug(f"Kokoro TTS provider not available: {e}")
|
35 |
+
|
36 |
+
# Try to register Dia provider
|
37 |
+
try:
|
38 |
+
from .dia_provider import DiaTTSProvider
|
39 |
+
self._providers['dia'] = DiaTTSProvider
|
40 |
+
logger.info("Registered Dia TTS provider")
|
41 |
+
except ImportError as e:
|
42 |
+
logger.debug(f"Dia TTS provider not available: {e}")
|
43 |
+
|
44 |
+
# Try to register CosyVoice2 provider
|
45 |
+
try:
|
46 |
+
from .cosyvoice2_provider import CosyVoice2TTSProvider
|
47 |
+
self._providers['cosyvoice2'] = CosyVoice2TTSProvider
|
48 |
+
logger.info("Registered CosyVoice2 TTS provider")
|
49 |
+
except ImportError as e:
|
50 |
+
logger.debug(f"CosyVoice2 TTS provider not available: {e}")
|
51 |
+
|
52 |
+
def get_available_providers(self) -> List[str]:
|
53 |
+
"""Get list of available TTS providers."""
|
54 |
+
available = []
|
55 |
+
for name, provider_class in self._providers.items():
|
56 |
+
try:
|
57 |
+
# Create instance if not cached
|
58 |
+
if name not in self._provider_instances:
|
59 |
+
if name == 'kokoro':
|
60 |
+
self._provider_instances[name] = provider_class()
|
61 |
+
elif name == 'dia':
|
62 |
+
self._provider_instances[name] = provider_class()
|
63 |
+
elif name == 'cosyvoice2':
|
64 |
+
self._provider_instances[name] = provider_class()
|
65 |
+
else:
|
66 |
+
self._provider_instances[name] = provider_class()
|
67 |
+
|
68 |
+
# Check if provider is available
|
69 |
+
if self._provider_instances[name].is_available():
|
70 |
+
available.append(name)
|
71 |
+
|
72 |
+
except Exception as e:
|
73 |
+
logger.warning(f"Failed to check availability of {name} provider: {e}")
|
74 |
+
|
75 |
+
return available
|
76 |
+
|
77 |
+
def create_provider(self, provider_name: str, **kwargs) -> TTSProviderBase:
|
78 |
+
"""
|
79 |
+
Create a TTS provider instance.
|
80 |
+
|
81 |
+
Args:
|
82 |
+
provider_name: Name of the provider to create
|
83 |
+
**kwargs: Additional arguments for provider initialization
|
84 |
+
|
85 |
+
Returns:
|
86 |
+
TTSProviderBase: The created provider instance
|
87 |
+
|
88 |
+
Raises:
|
89 |
+
SpeechSynthesisException: If provider is not available or creation fails
|
90 |
+
"""
|
91 |
+
if provider_name not in self._providers:
|
92 |
+
available = list(self._providers.keys())
|
93 |
+
raise SpeechSynthesisException(
|
94 |
+
f"Unknown TTS provider: {provider_name}. Available providers: {available}"
|
95 |
+
)
|
96 |
+
|
97 |
+
try:
|
98 |
+
provider_class = self._providers[provider_name]
|
99 |
+
|
100 |
+
# Create instance with appropriate parameters
|
101 |
+
if provider_name in ['kokoro', 'dia', 'cosyvoice2']:
|
102 |
+
lang_code = kwargs.get('lang_code', 'z')
|
103 |
+
provider = provider_class(lang_code=lang_code)
|
104 |
+
else:
|
105 |
+
provider = provider_class(**kwargs)
|
106 |
+
|
107 |
+
# Verify the provider is available
|
108 |
+
if not provider.is_available():
|
109 |
+
raise SpeechSynthesisException(f"TTS provider {provider_name} is not available")
|
110 |
+
|
111 |
+
logger.info(f"Created TTS provider: {provider_name}")
|
112 |
+
return provider
|
113 |
+
|
114 |
+
except Exception as e:
|
115 |
+
logger.error(f"Failed to create TTS provider {provider_name}: {e}")
|
116 |
+
raise SpeechSynthesisException(f"Failed to create TTS provider {provider_name}: {e}") from e
|
117 |
+
|
118 |
+
def get_provider_with_fallback(self, preferred_providers: List[str] = None, **kwargs) -> TTSProviderBase:
|
119 |
+
"""
|
120 |
+
Get a TTS provider with fallback logic.
|
121 |
+
|
122 |
+
Args:
|
123 |
+
preferred_providers: List of preferred providers in order of preference
|
124 |
+
**kwargs: Additional arguments for provider initialization
|
125 |
+
|
126 |
+
Returns:
|
127 |
+
TTSProviderBase: The first available provider
|
128 |
+
|
129 |
+
Raises:
|
130 |
+
SpeechSynthesisException: If no providers are available
|
131 |
+
"""
|
132 |
+
if preferred_providers is None:
|
133 |
+
preferred_providers = ['kokoro', 'dia', 'cosyvoice2', 'dummy']
|
134 |
+
|
135 |
+
available_providers = self.get_available_providers()
|
136 |
+
|
137 |
+
# Try preferred providers in order
|
138 |
+
for provider_name in preferred_providers:
|
139 |
+
if provider_name in available_providers:
|
140 |
+
try:
|
141 |
+
return self.create_provider(provider_name, **kwargs)
|
142 |
+
except Exception as e:
|
143 |
+
logger.warning(f"Failed to create preferred provider {provider_name}: {e}")
|
144 |
+
continue
|
145 |
+
|
146 |
+
# If no preferred providers work, try any available provider
|
147 |
+
for provider_name in available_providers:
|
148 |
+
if provider_name not in preferred_providers:
|
149 |
+
try:
|
150 |
+
return self.create_provider(provider_name, **kwargs)
|
151 |
+
except Exception as e:
|
152 |
+
logger.warning(f"Failed to create fallback provider {provider_name}: {e}")
|
153 |
+
continue
|
154 |
+
|
155 |
+
raise SpeechSynthesisException("No TTS providers are available")
|
156 |
+
|
157 |
+
def get_provider_info(self, provider_name: str) -> Dict:
|
158 |
+
"""
|
159 |
+
Get information about a specific provider.
|
160 |
+
|
161 |
+
Args:
|
162 |
+
provider_name: Name of the provider
|
163 |
+
|
164 |
+
Returns:
|
165 |
+
Dict: Provider information including availability and supported features
|
166 |
+
"""
|
167 |
+
if provider_name not in self._providers:
|
168 |
+
return {"available": False, "error": "Provider not registered"}
|
169 |
+
|
170 |
+
try:
|
171 |
+
# Create instance if not cached
|
172 |
+
if provider_name not in self._provider_instances:
|
173 |
+
provider_class = self._providers[provider_name]
|
174 |
+
if provider_name in ['kokoro', 'dia', 'cosyvoice2']:
|
175 |
+
self._provider_instances[provider_name] = provider_class()
|
176 |
+
else:
|
177 |
+
self._provider_instances[provider_name] = provider_class()
|
178 |
+
|
179 |
+
provider = self._provider_instances[provider_name]
|
180 |
+
|
181 |
+
return {
|
182 |
+
"available": provider.is_available(),
|
183 |
+
"name": provider.provider_name,
|
184 |
+
"supported_languages": provider.supported_languages,
|
185 |
+
"available_voices": provider.get_available_voices() if provider.is_available() else []
|
186 |
+
}
|
187 |
+
|
188 |
+
except Exception as e:
|
189 |
+
return {
|
190 |
+
"available": False,
|
191 |
+
"error": str(e)
|
192 |
+
}
|
193 |
+
|
194 |
+
def cleanup_providers(self):
|
195 |
+
"""Clean up provider instances and resources."""
|
196 |
+
for provider in self._provider_instances.values():
|
197 |
+
try:
|
198 |
+
if hasattr(provider, '_cleanup_temp_files'):
|
199 |
+
provider._cleanup_temp_files()
|
200 |
+
except Exception as e:
|
201 |
+
logger.warning(f"Failed to cleanup provider {provider.provider_name}: {e}")
|
202 |
+
|
203 |
+
self._provider_instances.clear()
|
204 |
+
logger.info("Cleaned up TTS provider instances")
|
test_tts_migration.py
ADDED
@@ -0,0 +1,137 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""Test script to verify TTS provider migration."""
|
3 |
+
|
4 |
+
import sys
|
5 |
+
import os
|
6 |
+
|
7 |
+
# Add src to path
|
8 |
+
sys.path.insert(0, os.path.join(os.path.dirname(__file__), 'src'))
|
9 |
+
|
10 |
+
def test_provider_imports():
|
11 |
+
"""Test that all providers can be imported."""
|
12 |
+
print("Testing provider imports...")
|
13 |
+
|
14 |
+
try:
|
15 |
+
from src.infrastructure.tts import TTSProviderFactory, DummyTTSProvider
|
16 |
+
print("β Core TTS components imported successfully")
|
17 |
+
except Exception as e:
|
18 |
+
print(f"β Failed to import core TTS components: {e}")
|
19 |
+
return False
|
20 |
+
|
21 |
+
try:
|
22 |
+
from src.domain.models.text_content import TextContent
|
23 |
+
from src.domain.models.voice_settings import VoiceSettings
|
24 |
+
from src.domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
25 |
+
print("β Domain models imported successfully")
|
26 |
+
except Exception as e:
|
27 |
+
print(f"β Failed to import domain models: {e}")
|
28 |
+
return False
|
29 |
+
|
30 |
+
return True
|
31 |
+
|
32 |
+
def test_dummy_provider():
|
33 |
+
"""Test the dummy provider functionality."""
|
34 |
+
print("\nTesting dummy provider...")
|
35 |
+
|
36 |
+
try:
|
37 |
+
from src.infrastructure.tts import DummyTTSProvider
|
38 |
+
from src.domain.models.text_content import TextContent
|
39 |
+
from src.domain.models.voice_settings import VoiceSettings
|
40 |
+
from src.domain.models.speech_synthesis_request import SpeechSynthesisRequest
|
41 |
+
|
42 |
+
# Create provider
|
43 |
+
provider = DummyTTSProvider()
|
44 |
+
print(f"β Created dummy provider: {provider.provider_name}")
|
45 |
+
|
46 |
+
# Check availability
|
47 |
+
if provider.is_available():
|
48 |
+
print("β Dummy provider is available")
|
49 |
+
else:
|
50 |
+
print("β Dummy provider is not available")
|
51 |
+
return False
|
52 |
+
|
53 |
+
# Check voices
|
54 |
+
voices = provider.get_available_voices()
|
55 |
+
print(f"β Available voices: {voices}")
|
56 |
+
|
57 |
+
# Create a synthesis request
|
58 |
+
text_content = TextContent(text="Hello, world!", language="en")
|
59 |
+
voice_settings = VoiceSettings(voice_id="default", speed=1.0, language="en")
|
60 |
+
request = SpeechSynthesisRequest(
|
61 |
+
text_content=text_content,
|
62 |
+
voice_settings=voice_settings
|
63 |
+
)
|
64 |
+
print("β Created synthesis request")
|
65 |
+
|
66 |
+
# Test synthesis
|
67 |
+
audio_content = provider.synthesize(request)
|
68 |
+
print(f"β Generated audio: {len(audio_content.data)} bytes, {audio_content.duration:.2f}s")
|
69 |
+
|
70 |
+
return True
|
71 |
+
|
72 |
+
except Exception as e:
|
73 |
+
print(f"β Dummy provider test failed: {e}")
|
74 |
+
import traceback
|
75 |
+
traceback.print_exc()
|
76 |
+
return False
|
77 |
+
|
78 |
+
def test_provider_factory():
|
79 |
+
"""Test the provider factory."""
|
80 |
+
print("\nTesting provider factory...")
|
81 |
+
|
82 |
+
try:
|
83 |
+
from src.infrastructure.tts import TTSProviderFactory
|
84 |
+
|
85 |
+
factory = TTSProviderFactory()
|
86 |
+
print("β Created provider factory")
|
87 |
+
|
88 |
+
available = factory.get_available_providers()
|
89 |
+
print(f"β Available providers: {available}")
|
90 |
+
|
91 |
+
if 'dummy' not in available:
|
92 |
+
print("β Dummy provider should always be available")
|
93 |
+
return False
|
94 |
+
|
95 |
+
# Test creating dummy provider
|
96 |
+
provider = factory.create_provider('dummy')
|
97 |
+
print(f"β Created provider via factory: {provider.provider_name}")
|
98 |
+
|
99 |
+
# Test fallback logic
|
100 |
+
provider = factory.get_provider_with_fallback()
|
101 |
+
print(f"β Got provider with fallback: {provider.provider_name}")
|
102 |
+
|
103 |
+
return True
|
104 |
+
|
105 |
+
except Exception as e:
|
106 |
+
print(f"β Provider factory test failed: {e}")
|
107 |
+
import traceback
|
108 |
+
traceback.print_exc()
|
109 |
+
return False
|
110 |
+
|
111 |
+
def main():
|
112 |
+
"""Run all tests."""
|
113 |
+
print("=== TTS Provider Migration Test ===\n")
|
114 |
+
|
115 |
+
tests = [
|
116 |
+
test_provider_imports,
|
117 |
+
test_dummy_provider,
|
118 |
+
test_provider_factory
|
119 |
+
]
|
120 |
+
|
121 |
+
passed = 0
|
122 |
+
for test in tests:
|
123 |
+
if test():
|
124 |
+
passed += 1
|
125 |
+
print()
|
126 |
+
|
127 |
+
print(f"=== Results: {passed}/{len(tests)} tests passed ===")
|
128 |
+
|
129 |
+
if passed == len(tests):
|
130 |
+
print("π All tests passed! TTS provider migration successful.")
|
131 |
+
return 0
|
132 |
+
else:
|
133 |
+
print("β Some tests failed. Check the output above.")
|
134 |
+
return 1
|
135 |
+
|
136 |
+
if __name__ == "__main__":
|
137 |
+
sys.exit(main())
|