File size: 5,708 Bytes
c318cb1 d347764 c318cb1 d347764 c318cb1 d347764 c318cb1 b9bb0a2 c318cb1 b9bb0a2 c318cb1 d347764 c318cb1 d347764 c318cb1 d347764 c318cb1 d347764 f805e49 b9bb0a2 f805e49 c737803 d347764 226ec3a d347764 f805e49 d347764 c737803 3946ba6 c737803 b9bb0a2 |
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 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 |
# import gradio as gr
# import numpy as np
# import torch
# from datasets import load_dataset
# from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
# device = "cuda:0" if torch.cuda.is_available() else "cpu"
# # load speech translation checkpoint
# asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device=device)
# # load text-to-speech checkpoint and speaker embeddings
# model_id = "microsoft/speecht5_tts" #"Ellight/speecht5_finetuned_voxpopuli_nl" # update with your model id
# # pipe = pipeline("automatic-speech-recognition", model=model_id)
# model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
# embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation",trust_remote_code=True))
# speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
# # speaker_embeddings = torch.tensor(embeddings_dataset[7440]["xvector"]).unsqueeze(0)
# processor = SpeechT5Processor.from_pretrained(model_id)
# replacements = [
# ("à", "a"),
# ("ç", "c"),
# ("è", "e"),
# ("ë", "e"),
# ("í", "i"),
# ("ï", "i"),
# ("ö", "o"),
# ("ü", "u"),
# ]
# def cleanup_text(text):
# for src, dst in replacements:
# text = text.replace(src, dst)
# return text
# def synthesize_speech(text):
# text = cleanup_text(text)
# inputs = processor(text=text, return_tensors="pt")
# speech = model.generate_speech(inputs["input_ids"].to(device), speaker_embeddings.to(device), vocoder=vocoder)
# return gr.Audio.update(value=(16000, speech.cpu().numpy()))
# def translate(audio):
# outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "Dutch"})
# return outputs["text"]
# def synthesise(text):
# text = cleanup_text(text)
# 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
import gradio as gr
import numpy as np
import torch
from datasets import load_dataset
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
from transformers import VitsModel, VitsTokenizer
device = "cuda:0" if torch.cuda.is_available() else "cpu"
# load speech translation checkpoint
asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
# load text-to-speech checkpoint and speaker embeddings
# processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
# model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
model = VitsModel.from_pretrained("Matthijs/mms-tts-nld")
tokenizer = VitsTokenizer.from_pretrained("Matthijs/mms-tts-nld")
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
def translate(audio):
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "transcribe", "language": "nl"})
return outputs["text"]
def synthesise(text):
inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
outputs = model(inputs["input_ids"])
speech = outputs.audio[0]
return speech.cpu()
# def synthesise(text):
# inputs = processor(text=text, return_tensors="pt", padding='max_length', truncation=True)
# 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 Dutch. Demo uses OpenAI's [Whisper Large v2](https://huggingface.co/openai/whisper-large-v2) model for speech translation, and [Sandiago21/speecht5_finetuned_voxpopuli_it](https://huggingface.co/Sandiago21/speecht5_finetuned_voxpopuli_it) checkpoint for text-to-speech, which is based on Microsoft's
[SpeechT5 TTS](https://huggingface.co/microsoft/speecht5_tts) model for text-to-speech, fine-tuned in Dutch Audio dataset:
"""
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() |