Spaces:
Runtime error
Runtime error
File size: 3,457 Bytes
d347764 f1d52a5 d347764 5285dde 37ec437 33fabd3 d347764 33fabd3 ca21522 d347764 ca21522 d347764 f1d52a5 5285dde f1d52a5 b667341 d347764 5285dde ca21522 f1d52a5 ca21522 d347764 ca21522 d347764 f805e49 c6f1d54 4f91edb f805e49 c737803 d347764 226ec3a d347764 f805e49 d347764 c737803 3946ba6 c737803 d347764 |
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 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 |
import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
from transformers import (
SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, WhisperForConditionalGeneration, WhisperFeatureExtractor, WhisperTokenizer, pipeline,
BarkModel, BarkProcessor
)
device = "cuda:0" if torch.cuda.is_available() else "cpu"
feature_extractor = WhisperFeatureExtractor.from_pretrained("openai/whisper-base")
tokenizer = WhisperTokenizer.from_pretrained("openai/whisper-base", language="french", task="translate")
whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-base")
forced_decoder_ids = tokenizer.get_decoder_prompt_ids(language="french", task="translate")
#load text-to-speech checkpoint and speaker embeddings
processor = SpeechT5Processor.from_pretrained("Apocalypse-19/speecht5_finetuned_french")
model = SpeechT5ForTextToSpeech.from_pretrained("Apocalypse-19/speecht5_finetuned_french").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
def translate(audio):
# load speech translation checkpoint
asr_pipe = pipeline(
"automatic-speech-recognition",
model=whisper_model,
feature_extractor=feature_extractor,
tokenizer=tokenizer,
device=device
)
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "forced_decoder_ids": forced_decoder_ids})
return outputs["text"]
def synthesise(text):
# inputs = processor(text, voice_preset="v2/fr_speaker_1")
# speech = bark_model.generate(**inputs).cpu()
inputs = processor(text=text, return_tensors="pt")
speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
return speech.cpu()
def speech_to_speech_translation(audio):
translated_text = translate(audio)
synthesised_speech = synthesise(translated_text)
synthesised_speech = (synthesised_speech.numpy() * 32767).astype(np.int16)
return 16000, synthesised_speech
title = "Cascaded STST"
description = """
Demo for cascaded speech-to-speech translation (STST), mapping from source speech in any language to target speech in English. Demo uses OpenAI's [Whisper Base](https://huggingface.co/openai/whisper-base) model for speech translation, and Microsoft's
[SpeechT5 TTS](https://huggingface.co/leo-kwan/speecht5_finetuned_voxpopuli_lt) model for text-to-speech:
"""
demo = gr.Blocks()
mic_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(source="microphone", type="filepath"),
outputs=gr.Audio(label="Generated Speech", type="numpy"),
title=title,
description=description,
)
file_translate = gr.Interface(
fn=speech_to_speech_translation,
inputs=gr.Audio(source="upload", type="filepath"),
outputs=gr.Audio(label="Generated Speech", type="numpy"),
examples=[["./example.wav"]],
title=title,
description=description,
)
with demo:
gr.TabbedInterface([mic_translate, file_translate], ["Microphone", "Audio File"])
demo.launch()
|