Spaces:
Runtime error
Runtime error
File size: 3,396 Bytes
feba911 1ecefeb feba911 e83f42b feba911 33b8d42 241ba79 8ce4e5c 33b8d42 241ba79 8ce4e5c 241ba79 33b8d42 241ba79 8ce4e5c 241ba79 8ce4e5c 33b8d42 241ba79 33b8d42 241ba79 33b8d42 feba911 |
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 |
import gradio as gr
#Get models
#ASR model for input speech
speech2text = gr.Interface.load("huggingface/facebook/hubert-large-ls960-ft",
inputs=gr.inputs.Audio(label="Record Audio File", type="file", source = "microphone"))
#translates english to spanish text
translator = gr.Interface.load("huggingface/Helsinki-NLP/opus-mt-en-es",
outputs=gr.outputs.Textbox(label="English to Spanish Translated Text"))
#TTS model for output speech
text2speech = gr.Interface.load("huggingface/facebook/tts_transformer-es-css10",
outputs=gr.outputs.Audio(label="English to Spanish Translated Audio"),
allow_flagging="never")
translate = gr.Series(speech2text, translator) #outputs Spanish text translation
en2es = gr.Series(translate, text2speech) #outputs Spanish audio
ui = gr.Parallel(translate, en2es) #allows transcription of Spanish audio
#gradio interface
ui.title = "English to Spanish Speech Translator"
ui.description = """<center>A useful tool in translating English to Spanish audio. All pre-trained models are found in huggingface.</center>"""
ui.examples = [['ljspeech.wav'],['ljspeech2.wav',]]
ui.theme = "peach"
ui.article = """<h2>Pre-trained model Information</h2>
<h3>Automatic Speech Recognition</h3>
<p style='text-align: justify'>The model used for the ASR part of this space is from
[https://huggingface.co/facebook/wav2vec2-base-960h] which is pretrained and fine-tuned on
<b>960 hours of
Librispeech</b> on 16kHz sampled speech audio. This model has a <b>word error rate (WER)</b> of <b>8.6 percent on
noisy speech</b> and <b>5.2 percent on clean speech</b> on the standard LibriSpeech benchmark. More information can be
found on its website at [https://ai.facebook.com/blog/wav2vec-20-learning-the-structure-of-speech-from-raw-audio]
and
original model is under [https://github.com/pytorch/fairseq/tree/main/examples/wav2vec].</p>
<h3>Text Translator</h3>
<p style='text-align: justify'>The English to Spanish text translator pre-trained model is from
[https://huggingface.co/Helsinki-NLP/opus-mt-en-es] which is part of the <b>The Tatoeba Translation Challenge
(v2021-08-07)</b> as seen from its github repo at
[https://github.com/Helsinki-NLP/Tatoeba-Challenge]. This project aims to develop machine
translation in real-world
cases for many languages. </p>
<h3>Text to Speech</h3>
<p style='text-align: justify'> The TTS model used is from [https://huggingface.co/facebook/tts_transformer-es-css10].
This model uses the <b>Fairseq(-py)</b> sequence modeling toolkit for speech synthesis, in this case, specifically TTS
for Spanish. More information can be seen on their git at
[https://github.com/pytorch/fairseq/tree/main/examples/speech_synthesis]. </p>
"""
ui.launch(inbrowser=True)
|