File size: 4,432 Bytes
785694e 28456aa 785694e 8e59fb5 785694e 6fdf224 785694e 6fdf224 785694e 6fdf224 785694e 6fdf224 785694e 3316cf6 6fdf224 8e59fb5 6fdf224 3316cf6 6fdf224 3316cf6 6fdf224 785694e 6fdf224 c3d13cb 785694e 09896ad 785694e e701653 785694e e701653 785694e |
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 |
import gradio as gr
import numpy as np
import os
import torch
from datasets import load_dataset
from transformers import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor, pipeline
from speechbrain.pretrained import EncoderClassifier
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("microsoft/speecht5_tts").to(device)
# vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan").to(device)
processor = SpeechT5Processor.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl")
model = SpeechT5ForTextToSpeech.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
vocoder = SpeechT5HifiGan.from_pretrained("sanchit-gandhi/speecht5_tts_vox_nl").to(device)
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
spk_model_name = "speechbrain/spkrec-xvect-voxceleb"
device = "cuda" if torch.cuda.is_available() else "cpu"
speaker_model = EncoderClassifier.from_hparams(
source=spk_model_name,
run_opts={"device": device},
savedir=os.path.join("/tmp", spk_model_name),
)
def create_speaker_embedding(waveform):
with torch.no_grad():
speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform))
speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=2)
speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
return speaker_embeddings
dataset_nl = load_dataset("facebook/voxpopuli", "nl", split="train", streaming=True)
data_list = []
speaker_embeddings_list = []
for i, data in enumerate(iter(dataset_nl)):
# print(i)
if(i > 16):
break
data_list.append(data)
# data = next(iter(dataset_nl))
text = data["raw_text"]
# print(data)
speaker_embeddings = create_speaker_embedding(data["audio"]["array"])
speaker_embeddings = torch.tensor(speaker_embeddings)[None]
speaker_embeddings_list.append(speaker_embeddings)
speaker_embeddings = speaker_embeddings_list[11]
def translate(audio):
# outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"task": "translate"})
outputs = asr_pipe(audio, max_new_tokens=256, generate_kwargs={"language":"<|nl|>","task": "transcribe"})
return outputs["text"]
def synthesise(text):
#inputs = processor(text=text, return_tensors="pt")
inputs = processor(text=text, return_tensors="pt", truncation=True, max_length=200)
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)
print(translated_text)
synthesised_speech = synthesise(translated_text)
print(synthesised_speech)
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/microsoft/speecht5_tts) 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()
|