Spaces:
Build error
Build error
File size: 2,557 Bytes
41ad754 ff017ef 41ad754 aad7d40 41ad754 ff017ef 41ad754 aad7d40 41ad754 aad7d40 41ad754 aad7d40 ff017ef aad7d40 1064862 41ad754 aad7d40 41ad754 1064862 41ad754 ff017ef aad7d40 ff017ef 41ad754 aad7d40 41ad754 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 |
import gradio as gr
import os
import torch
import soundfile as sf
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor, VitsModel, AutoTokenizer
# Ensure the output directory exists
os.makedirs("output_audio", exist_ok=True)
# Load the models and processors
asr_processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-960h")
asr_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h")
tts_model = VitsModel.from_pretrained("facebook/mms-tts-eng")
tts_tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")
def speech_to_text(input_audio):
# Load and preprocess the audio
waveform, sr = sf.read(input_audio)
input_values = asr_processor(waveform, sampling_rate=sr, return_tensors="pt").input_values
# Perform speech recognition
with torch.no_grad():
logits = asr_model(input_values).logits
predicted_ids = torch.argmax(logits, dim=-1)
# Decode the predicted IDs to text
transcription = asr_processor.batch_decode(predicted_ids)[0]
return transcription
def text_to_speech(text):
# Tokenize text and generate waveform
inputs = tts_tokenizer(text, return_tensors="pt")
with torch.no_grad():
output = tts_model(**inputs).waveform
waveform = output.numpy()
# Define output path and save waveform as audio file
output_path = "output_audio/text_to_speech.wav"
sf.write(output_path, waveform.squeeze(), 22050)
return output_path
def speech_to_speech(input_audio, target_text):
# Synthesize speech directly from target text without transcribing the input audio
return text_to_speech(target_text)
iface = gr.Interface(
fn={
"Speech to Text": speech_to_text,
"Text to Speech": text_to_speech,
"Speech to Speech": speech_to_speech
},
inputs={
"Speech to Text": gr.inputs.Audio(source="upload", type="file"),
"Text to Speech": gr.inputs.Textbox(label="Text"),
"Speech to Speech": [gr.inputs.Audio(source="upload", type="file"), gr.inputs.Textbox(label="Target Text")]
},
outputs={
"Speech to Text": gr.outputs.Textbox(label="Transcription"),
"Text to Speech": gr.outputs.Audio(type="file", label="Synthesized Speech"),
"Speech to Speech": gr.outputs.Audio(type="file", label="Synthesized Speech")
},
title="Speech Processing Application",
description="This app uses Facebook's Wav2Vec 2.0 for speech-to-text and VITS for text-to-speech.",
layout="vertical"
)
if __name__ == "__main__":
iface.launch()
|