Pocket-TTS / app.py
FresherDifference's picture
Update app.py
909f184 verified
import gradio as gr
import numpy as np
import os
from pydub import AudioSegment
from pocket_tts import TTSModel
# 1. Load the model
print("Loading Pocket-TTS model...")
# Ensure you have HF_TOKEN in your Space Secrets for cloning to work
tts = TTSModel.load_model()
print("Model loaded successfully.")
# 2. Define Presets (Simple Strings Only)
# We map the display name to the internal ID string.
# We do NOT use URLs here to avoid 404 errors.
PRESET_VOICES = {
"Alba (American English)": "alba",
"Marius (French)": "marius",
"Jean (Narrator)": "jean",
"Fantine": "fantine",
"Javert": "javert",
"Cosette": "cosette",
"Eponine": "eponine",
"Azelma": "azelma",
}
def preprocess_audio(filepath):
"""
Takes any audio file (MP3, M4A, WAV), ensures it is
a valid 16-bit PCM WAV compatible with the model.
"""
try:
print(f"Converting file: {filepath}")
audio = AudioSegment.from_file(filepath)
# Pocket-TTS works best with mono, 24000Hz or 16000Hz, 16-bit
# We enforce standard wav settings here to prevent "RIFF id" errors
audio = audio.set_channels(1).set_sample_width(2)
output_path = filepath + "_fixed.wav"
audio.export(output_path, format="wav")
print(f"Converted to: {output_path}")
return output_path
except Exception as e:
raise gr.Error(f"Audio conversion failed. Make sure ffmpeg is installed in packages.txt. Error: {e}")
def generate_speech(text, voice_choice, custom_voice_file):
if not text.strip():
raise gr.Error("Please enter some text.")
try:
# LOGIC BRANCH 1: Custom Voice Upload
if custom_voice_file is not None:
print("--- Mode: Voice Cloning ---")
# 1. Fix the audio file (Fixes 'RIFF id' error)
clean_wav_path = preprocess_audio(custom_voice_file)
# 2. Extract the speaker style
# The model analyzes the WAV to clone the voice
voice_state = tts.get_state_for_audio_prompt(clean_wav_path)
# 3. Generate
audio_tensor = tts.generate_audio(voice_state, text)
# LOGIC BRANCH 2: Built-in Preset
else:
print("--- Mode: Preset Voice ---")
voice_id = PRESET_VOICES[voice_choice]
print(f"Using Internal ID: {voice_id}")
# We pass the STRING directly.
# We do NOT use get_state_for_audio_prompt for presets (Fixes 404 error)
audio_tensor = tts.generate_audio(voice_id, text)
return (tts.sample_rate, audio_tensor.numpy())
except Exception as e:
# Print full error to logs for debugging
import traceback
traceback.print_exc()
raise gr.Error(f"Generation Error: {str(e)}")
# 3. Build Interface
with gr.Blocks(title="Pocket-TTS Demo") as demo:
gr.Markdown("# 🗣️ Pocket-TTS (Fixed)")
gr.Markdown("Supports Voice Cloning (MP3/WAV) & Built-in Voices.")
with gr.Row():
with gr.Column():
text_input = gr.Textbox(
label="Text to Speak",
lines=4,
value="This is a test of the pocket text to speech system."
)
with gr.Accordion("Voice Settings", open=True):
voice_dropdown = gr.Dropdown(
choices=list(PRESET_VOICES.keys()),
value="Alba (American English)",
label="Use a Preset Voice"
)
gr.Markdown("**OR**")
voice_upload = gr.Audio(
label="Clone a Voice (Upload any audio)",
type="filepath"
)
submit_btn = gr.Button("Generate Audio", variant="primary")
with gr.Column():
audio_output = gr.Audio(label="Result", type="numpy")
submit_btn.click(
fn=generate_speech,
inputs=[text_input, voice_dropdown, voice_upload],
outputs=audio_output
)
demo.launch()