Commit
·
49b2f3e
1
Parent(s):
17f4c83
Add subtitle generation feature to text-to-speech functionality
Browse files
app.py
CHANGED
@@ -3,6 +3,8 @@ import edge_tts
|
|
3 |
import asyncio
|
4 |
import tempfile
|
5 |
import os
|
|
|
|
|
6 |
|
7 |
|
8 |
async def get_voices():
|
@@ -13,11 +15,21 @@ async def get_voices():
|
|
13 |
}
|
14 |
|
15 |
|
16 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
if not text.strip():
|
18 |
-
return None, "Please enter text to convert."
|
19 |
if not voice:
|
20 |
-
return None, "Please select a voice."
|
21 |
|
22 |
voice_short_name = voice.split(" - ")[0]
|
23 |
rate_str = f"{rate:+d}%"
|
@@ -25,17 +37,95 @@ async def text_to_speech(text, voice, rate, pitch):
|
|
25 |
communicate = edge_tts.Communicate(
|
26 |
text, voice_short_name, rate=rate_str, pitch=pitch_str
|
27 |
)
|
|
|
|
|
28 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
|
29 |
-
|
30 |
-
|
31 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
|
33 |
|
34 |
-
async def tts_interface(text, voice, rate, pitch):
|
35 |
-
audio, warning = await text_to_speech(text, voice, rate, pitch)
|
36 |
if warning:
|
37 |
-
return audio, gr.Warning(warning)
|
38 |
-
return audio, None
|
39 |
|
40 |
|
41 |
async def create_demo():
|
@@ -43,6 +133,7 @@ async def create_demo():
|
|
43 |
|
44 |
description = """
|
45 |
Convert text to speech using Microsoft Edge TTS. Adjust speech rate and pitch: 0 is default, positive values increase, negative values decrease.
|
|
|
46 |
|
47 |
**Note:** Edge TTS is a cloud-based service and requires an active internet connection."""
|
48 |
|
@@ -65,9 +156,11 @@ async def create_demo():
|
|
65 |
gr.Slider(
|
66 |
minimum=-20, maximum=20, value=0, label="Pitch Adjustment (Hz)", step=1
|
67 |
),
|
|
|
68 |
],
|
69 |
outputs=[
|
70 |
gr.Audio(label="Generated Audio", type="filepath"),
|
|
|
71 |
gr.Markdown(label="Warning", visible=False),
|
72 |
],
|
73 |
title="Edge TTS Text-to-Speech",
|
|
|
3 |
import asyncio
|
4 |
import tempfile
|
5 |
import os
|
6 |
+
import json
|
7 |
+
import datetime
|
8 |
|
9 |
|
10 |
async def get_voices():
|
|
|
15 |
}
|
16 |
|
17 |
|
18 |
+
def format_time(milliseconds):
|
19 |
+
"""Convert milliseconds to SRT time format (HH:MM:SS,mmm)"""
|
20 |
+
# Ensure milliseconds is an integer
|
21 |
+
milliseconds = int(milliseconds)
|
22 |
+
seconds, milliseconds = divmod(milliseconds, 1000)
|
23 |
+
minutes, seconds = divmod(seconds, 60)
|
24 |
+
hours, minutes = divmod(minutes, 60)
|
25 |
+
return f"{hours:02d}:{minutes:02d}:{seconds:02d},{milliseconds:03d}"
|
26 |
+
|
27 |
+
|
28 |
+
async def text_to_speech(text, voice, rate, pitch, generate_subtitles=False):
|
29 |
if not text.strip():
|
30 |
+
return None, None, "Please enter text to convert."
|
31 |
if not voice:
|
32 |
+
return None, None, "Please select a voice."
|
33 |
|
34 |
voice_short_name = voice.split(" - ")[0]
|
35 |
rate_str = f"{rate:+d}%"
|
|
|
37 |
communicate = edge_tts.Communicate(
|
38 |
text, voice_short_name, rate=rate_str, pitch=pitch_str
|
39 |
)
|
40 |
+
|
41 |
+
# Create temporary file for audio
|
42 |
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
|
43 |
+
audio_path = tmp_file.name
|
44 |
+
|
45 |
+
subtitle_path = None
|
46 |
+
if generate_subtitles:
|
47 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".srt") as srt_file:
|
48 |
+
subtitle_path = srt_file.name
|
49 |
+
|
50 |
+
# Generate audio and collect word boundary data
|
51 |
+
async def process_audio():
|
52 |
+
word_boundaries = []
|
53 |
+
async for chunk in communicate.stream():
|
54 |
+
if chunk["type"] == "audio":
|
55 |
+
with open(audio_path, "ab") as audio_file:
|
56 |
+
audio_file.write(chunk["data"])
|
57 |
+
elif chunk["type"] == "WordBoundary":
|
58 |
+
word_boundaries.append(chunk)
|
59 |
+
return word_boundaries
|
60 |
+
|
61 |
+
word_boundaries = await process_audio()
|
62 |
+
|
63 |
+
# Group words into sensible phrases/sentences for subtitles
|
64 |
+
phrases = []
|
65 |
+
current_phrase = []
|
66 |
+
current_text = ""
|
67 |
+
phrase_start = 0
|
68 |
+
|
69 |
+
for i, boundary in enumerate(word_boundaries):
|
70 |
+
word = boundary["text"]
|
71 |
+
start_time = boundary["offset"] / 10000
|
72 |
+
duration = boundary["duration"] / 10000
|
73 |
+
end_time = start_time + duration
|
74 |
+
|
75 |
+
if not current_phrase:
|
76 |
+
phrase_start = start_time
|
77 |
+
|
78 |
+
current_phrase.append(boundary)
|
79 |
+
current_text += word + " "
|
80 |
+
|
81 |
+
# Determine if we should end this phrase and start a new one
|
82 |
+
should_break = False
|
83 |
+
|
84 |
+
# Break on punctuation
|
85 |
+
if word.endswith(('.', '!', '?', ':', ';', ',')) or i == len(word_boundaries) - 1:
|
86 |
+
should_break = True
|
87 |
+
|
88 |
+
# Break after a certain number of words (4-5 is typical for subtitles)
|
89 |
+
elif len(current_phrase) >= 5:
|
90 |
+
should_break = True
|
91 |
+
|
92 |
+
# Break on long pause (more than 300ms between words)
|
93 |
+
elif i < len(word_boundaries) - 1:
|
94 |
+
next_start = word_boundaries[i + 1]["offset"] / 10000
|
95 |
+
if next_start - end_time > 300:
|
96 |
+
should_break = True
|
97 |
+
|
98 |
+
if should_break or i == len(word_boundaries) - 1:
|
99 |
+
if current_phrase:
|
100 |
+
last_boundary = current_phrase[-1]
|
101 |
+
phrase_end = (last_boundary["offset"] + last_boundary["duration"]) / 10000
|
102 |
+
phrases.append({
|
103 |
+
"text": current_text.strip(),
|
104 |
+
"start": phrase_start,
|
105 |
+
"end": phrase_end
|
106 |
+
})
|
107 |
+
current_phrase = []
|
108 |
+
current_text = ""
|
109 |
+
|
110 |
+
# Write phrases to SRT file
|
111 |
+
with open(subtitle_path, "w", encoding="utf-8") as srt_file:
|
112 |
+
for i, phrase in enumerate(phrases):
|
113 |
+
# Write SRT entry
|
114 |
+
srt_file.write(f"{i+1}\n")
|
115 |
+
srt_file.write(f"{format_time(phrase['start'])} --> {format_time(phrase['end'])}\n")
|
116 |
+
srt_file.write(f"{phrase['text']}\n\n")
|
117 |
+
else:
|
118 |
+
# Just generate audio
|
119 |
+
await communicate.save(audio_path)
|
120 |
+
|
121 |
+
return audio_path, subtitle_path, None
|
122 |
|
123 |
|
124 |
+
async def tts_interface(text, voice, rate, pitch, generate_subtitles):
|
125 |
+
audio, subtitle, warning = await text_to_speech(text, voice, rate, pitch, generate_subtitles)
|
126 |
if warning:
|
127 |
+
return audio, subtitle, gr.Warning(warning)
|
128 |
+
return audio, subtitle, None
|
129 |
|
130 |
|
131 |
async def create_demo():
|
|
|
133 |
|
134 |
description = """
|
135 |
Convert text to speech using Microsoft Edge TTS. Adjust speech rate and pitch: 0 is default, positive values increase, negative values decrease.
|
136 |
+
You can also generate subtitle files (.srt) along with the audio.
|
137 |
|
138 |
**Note:** Edge TTS is a cloud-based service and requires an active internet connection."""
|
139 |
|
|
|
156 |
gr.Slider(
|
157 |
minimum=-20, maximum=20, value=0, label="Pitch Adjustment (Hz)", step=1
|
158 |
),
|
159 |
+
gr.Checkbox(label="Generate Subtitles (.srt)", value=False),
|
160 |
],
|
161 |
outputs=[
|
162 |
gr.Audio(label="Generated Audio", type="filepath"),
|
163 |
+
gr.File(label="Generated Subtitles"),
|
164 |
gr.Markdown(label="Warning", visible=False),
|
165 |
],
|
166 |
title="Edge TTS Text-to-Speech",
|