zach
Restore encapsulation for ElevenLabs integration, update TTS functions to return the voice name in addition to the audio
36b195f
raw
history blame
15.5 kB
"""
app.py
Gradio UI for interacting with the Anthropic API, Hume TTS API, and ElevenLabs TTS API.
Users enter a prompt, which is processed using Claude by Anthropic to generate text.
The text is then synthesized into speech using both Hume and ElevenLabs text-to-speech (TTS) APIs.
Users can compare the outputs and vote for their favorite in an interactive UI.
"""
# Standard Library Imports
from concurrent.futures import ThreadPoolExecutor
import random
from typing import Union, Tuple
# Third-Party Library Imports
import gradio as gr
# Local Application Imports
from src.config import logger
from src.constants import (
ELEVENLABS,
HUME_AI,
OPTION_A,
OPTION_B,
PROMPT_MAX_LENGTH,
PROMPT_MIN_LENGTH,
SAMPLE_PROMPTS,
TROPHY_EMOJI,
UNKNOWN_PROVIDER,
VOTE_FOR_OPTION_A,
VOTE_FOR_OPTION_B,
)
from src.integrations import (
AnthropicError,
ElevenLabsError,
generate_text_with_claude,
get_random_hume_voice_names,
HumeError,
text_to_speech_with_elevenlabs,
text_to_speech_with_hume,
)
from src.theme import CustomTheme
from src.utils import truncate_text, validate_prompt_length
def generate_text(prompt: str,) -> Tuple[Union[str, gr.update], gr.update]:
"""
Validates the prompt and generates text using Anthropic API.
Args:
prompt (str): The user-provided text prompt.
Returns:
Tuple containing:
- The generated text (as a gr.update).
- An update for the generated text state.
Raises:
gr.Error: On validation or API errors.
"""
try:
validate_prompt_length(prompt, PROMPT_MAX_LENGTH, PROMPT_MIN_LENGTH)
except ValueError as ve:
logger.warning(f'Validation error: {ve}')
raise gr.Error(str(ve))
try:
generated_text = generate_text_with_claude(prompt)
logger.info(f'Generated text ({len(generated_text)} characters).')
return gr.update(value=generated_text), generated_text
except AnthropicError as ae:
logger.error(f'AnthropicError while generating text: {str(ae)}')
raise gr.Error('There was an issue communicating with the Anthropic API. Please try again later.')
except Exception as e:
logger.error(f'Unexpected error while generating text: {e}')
raise gr.Error('Failed to generate text. Please try again.')
def text_to_speech(prompt: str, text: str, generated_text_state: str) -> Tuple[gr.update, gr.update, dict, Union[str, None]]:
"""
Synthesizes two text to speech outputs, loads the two audio players with the
output audio, and updates related UI state components.
- 50% chance to synthesize one Hume and one Elevenlabs output.
- 50% chance to synthesize two Hume outputs.
Args:
prompt (str): The original prompt.
text (str): The text to synthesize to speech.
Returns:
A tuple of:
- Update for first audio player (with autoplay)
- Update for second audio player
- A dictionary mapping options to providers
- The raw audio value for option B
Raises:
gr.Error: On API or unexpected errors.
"""
if not text:
logger.warning('Skipping text-to-speech due to empty text.')
raise gr.Error('Please generate or enter text to synthesize.')
# If not using generated text, then only compare Hume to Hume
compare_hume_with_elevenlabs = (text == generated_text_state) and (random.random() < 0.5)
# Pre-select two Hume voices pre-emptively in case we compare Hume to Hume to ensure we do not select the same voice twice.
hume_voice_a, hume_voice_b = get_random_hume_voice_names()
try:
with ThreadPoolExecutor(max_workers=2) as executor:
provider_a = HUME_AI
future_audio_a = executor.submit(text_to_speech_with_hume, prompt, text, hume_voice_a)
if compare_hume_with_elevenlabs:
provider_b = ELEVENLABS
future_audio_b = executor.submit(text_to_speech_with_elevenlabs, text)
else:
provider_b = HUME_AI
future_audio_b = executor.submit(text_to_speech_with_hume, prompt, text, hume_voice_b)
voice_a, audio_a = future_audio_a.result()
voice_b, audio_b = future_audio_b.result()
logger.info(f'TTS generated: {provider_a}={len(audio_a)} bytes, {provider_b}={len(audio_b)} bytes')
options = [(audio_a, provider_a), (audio_b, provider_b)]
random.shuffle(options)
option_a_audio, option_b_audio = options[0][0], options[1][0]
options_map = { OPTION_A: options[0][1], OPTION_B: options[1][1] }
return (
gr.update(value=option_a_audio, visible=True, autoplay=True),
gr.update(value=option_b_audio, visible=True),
options_map,
option_b_audio,
)
except ElevenLabsError as ee:
logger.error(f'ElevenLabsError while synthesizing speech from text: {str(ee)}')
raise gr.Error('There was an issue communicating with the Elevenlabs API. Please try again later.')
except HumeError as he:
logger.error(f'HumeError while synthesizing speech from text: {str(he)}')
raise gr.Error('There was an issue communicating with the Hume API. Please try again later.')
except Exception as e:
logger.error(f'Unexpected error during TTS generation: {e}')
raise gr.Error('An unexpected error ocurred. Please try again later.')
def vote(vote_submitted: bool, option_mapping: dict, selected_button: str) -> Tuple[bool, gr.update, gr.update]:
"""
Handles user voting.
Args:
vote_submitted (bool): True if a vote was already submitted
option_mapping (dict): Maps option labels to provider names
selected_button (str): The button clicked
Returns:
A tuple of:
- True if the vote was accepted
- Update for the selected vote button
- Update for the unselected vote button
"""
if not option_mapping or vote_submitted:
return gr.skip(), gr.skip(), gr.skip()
is_option_a = selected_button == VOTE_FOR_OPTION_A
selected_option, other_option = (OPTION_A, OPTION_B) if is_option_a else (OPTION_B, OPTION_A)
selected_provider = option_mapping.get(selected_option, UNKNOWN_PROVIDER)
other_provider = option_mapping.get(other_option, UNKNOWN_PROVIDER)
return (
True,
gr.update(value=f'{selected_provider} {TROPHY_EMOJI}', variant='primary', interactive=False) if is_option_a
else gr.update(value=other_provider, variant='secondary', interactive=False),
gr.update(value=other_provider, variant='secondary', interactive=False) if is_option_a
else gr.update(value=f'{selected_provider} {TROPHY_EMOJI}', variant='primary', interactive=False),
gr.update(interactive=True)
)
def reset_ui() -> Tuple[gr.update, gr.update, gr.update, gr.update, None, None, bool]:
"""
Resets UI state before generating new text.
Returns:
A tuple of updates for:
- option_a_audio_player (clear audio)
- option_b_audio_player (clear audio)
- vote_button_a (disable and reset button text)
- vote_button_a (disable and reset button text)
- option_mapping_state (reset option map state)
- option_b_audio_state (reset option B audio state)
- vote_submitted_state (reset submitted vote state)
"""
return (
gr.update(value=None),
gr.update(value=None),
gr.update(value=VOTE_FOR_OPTION_A, variant='secondary'),
gr.update(value=VOTE_FOR_OPTION_B, variant='secondary'),
None,
None,
False,
)
def build_input_section() -> Tuple[gr.Markdown, gr.Dropdown, gr.Textbox, gr.Button]:
""" Builds the input section including instructions, sample prompt dropdown, prompt input, and generate button """
instructions = gr.Markdown("""
1. **Enter or Generate Text:** Type directly in the Text box, or optionally enter a Prompt, click "Generate text", and edit if needed.
2. **Synthesize Speech:** Click "Synthesize speech" to generate two audio outputs.
3. **Listen & Compare:** Playback both options (A & B) to hear the differences.
4. **Vote for Your Favorite:** Click "Vote for option A" or "Vote for option B" to choose the best one.
""")
sample_prompt_dropdown = gr.Dropdown(
choices=list(SAMPLE_PROMPTS.keys()),
label='Choose a sample prompt (or enter your own)',
value=None,
interactive=True,
)
prompt_input = gr.Textbox(
label='Prompt',
placeholder='Enter your prompt...',
lines=2,
max_lines=2,
max_length=PROMPT_MAX_LENGTH,
show_copy_button=True,
)
generate_text_button = gr.Button('Generate text', variant='secondary')
return (
instructions,
sample_prompt_dropdown,
prompt_input,
generate_text_button,
)
def build_output_section() -> Tuple[gr.Textbox, gr.Button, gr.Audio, gr.Audio, gr.Button, gr.Button]:
""" Builds the output section including generated text, audio players, and vote buttons. """
text_input = gr.Textbox(
label='Text',
placeholder='Enter text to synthesize speech...',
interactive=True,
autoscroll=False,
lines=4,
max_lines=12,
max_length=PROMPT_MAX_LENGTH,
show_copy_button=True,
)
synthesize_speech_button = gr.Button('Synthesize speech', variant='primary')
with gr.Row(equal_height=True):
option_a_audio_player = gr.Audio(label=OPTION_A, type='filepath', interactive=False)
option_b_audio_player = gr.Audio(label=OPTION_B, type='filepath', interactive=False)
with gr.Row(equal_height=True):
vote_button_a = gr.Button(VOTE_FOR_OPTION_A, interactive=False)
vote_button_b = gr.Button(VOTE_FOR_OPTION_B, interactive=False)
return (
text_input,
synthesize_speech_button,
option_a_audio_player,
option_b_audio_player,
vote_button_a,
vote_button_b,
)
def build_gradio_interface() -> gr.Blocks:
"""
Builds and configures the Gradio user interface.
Returns:
gr.Blocks: The fully constructed Gradio UI layout.
"""
custom_theme = CustomTheme()
with gr.Blocks(
title='Expressive TTS Arena',
theme=custom_theme,
fill_width=True,
css_paths='src/assets/styles.css',
) as demo:
# Title
gr.Markdown('# Expressive TTS Arena')
# Build generate text section
(instructions,
sample_prompt_dropdown,
prompt_input,
generate_text_button) = build_input_section()
# Build synthesize speech section
(text_input,
synthesize_speech_button,
option_a_audio_player,
option_b_audio_player,
vote_button_a,
vote_button_b) = build_output_section()
# UI state components
generated_text_state = gr.State('') # Track generated text state
option_b_audio_state = gr.State() # Track generated audio for option B for playing automatically after option 1 audio finishes
option_mapping_state = gr.State() # Track option map (option A and option B are randomized)
vote_submitted_state = gr.State(False) # Track whether the user has voted on an option
# --- Register event handlers ---
# When a sample prompt is chosen, update the prompt textbox
sample_prompt_dropdown.change(
fn=lambda choice: SAMPLE_PROMPTS.get(choice, ''),
inputs=[sample_prompt_dropdown],
outputs=[prompt_input],
)
# Generate text button click handler chain:
# 1. Disable the "Generate text" button
# 2. Generate text
# 3. Enable the "Generate text" button
generate_text_button.click(
fn=lambda: gr.update(interactive=False),
inputs=[],
outputs=[generate_text_button]
).then(
fn=generate_text,
inputs=[prompt_input],
outputs=[text_input, generated_text_state],
).then(
fn=lambda: gr.update(interactive=True),
inputs=[],
outputs=[generate_text_button]
)
# Synthesize speech button click event handler chain:
# 1. Disable the "Synthesize speech" button
# 2. Reset UI state
# 3. Synthesize speech, load audio players, and display vote button
# 4. Enable the "Synthesize speech" button and display vote buttons
synthesize_speech_button.click(
fn=lambda: gr.update(interactive=False),
inputs=[],
outputs=[synthesize_speech_button]
).then(
fn=reset_ui,
inputs=[],
outputs=[
option_a_audio_player,
option_b_audio_player,
vote_button_a,
vote_button_b,
option_mapping_state,
option_b_audio_state,
vote_submitted_state,
],
).then(
fn=text_to_speech,
inputs=[prompt_input, text_input, generated_text_state],
outputs=[
option_a_audio_player,
option_b_audio_player,
option_mapping_state,
option_b_audio_state,
],
).then(
fn=lambda: (
gr.update(interactive=True),
gr.update(interactive=True),
gr.update(interactive=True)
),
inputs=[],
outputs=[synthesize_speech_button, vote_button_a, vote_button_b]
)
# Vote button click event handlers
vote_button_a.click(
fn=vote,
inputs=[vote_submitted_state, option_mapping_state, vote_button_a],
outputs=[vote_submitted_state, vote_button_a, vote_button_b, synthesize_speech_button],
)
vote_button_b.click(
fn=vote,
inputs=[vote_submitted_state, option_mapping_state, vote_button_b],
outputs=[vote_submitted_state, vote_button_a, vote_button_b, synthesize_speech_button],
)
# Auto-play second audio after first finishes (Workaround to play audio back-to-back)
# Audio player A stop event handler chain:
# 1. Clear the audio player A
# 2. Load audio player A with audio and set auto play to True
option_a_audio_player.stop(
fn=lambda _: gr.update(value=None),
inputs=[],
outputs=[option_b_audio_player],
).then(
fn=lambda audio: gr.update(value=audio, autoplay=True),
inputs=[option_b_audio_state],
outputs=[option_b_audio_player],
)
# Enable voting after second audio option playback finishes
option_b_audio_player.stop(
fn=lambda _: (gr.update(interactive=True), gr.update(interactive=True), gr.update(autoplay=False)),
inputs=[],
outputs=[vote_button_a, vote_button_b, option_b_audio_player],
)
logger.debug('Gradio interface built successfully')
return demo
if __name__ == '__main__':
logger.info('Launching TTS Arena Gradio app...')
demo = build_gradio_interface()
demo.launch()