File size: 1,636 Bytes
9fd9429
f7a3b8e
 
 
9fd9429
3d50426
3eaab4f
f7a3b8e
90c0a2a
f7a3b8e
 
 
 
3a850f5
f7a3b8e
 
 
 
c5e7677
 
 
 
 
 
 
4c4b37d
 
f7a3b8e
2634b87
f7a3b8e
4c4b37d
0e902a4
4c4b37d
f7a3b8e
 
 
 
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
import os
os.system("pip install gradio==3.3")
import gradio as gr
import numpy as np
import streamlit as st

title = "Fairseq Speech to Speech Translation"

description = "Gradio Demo for fairseq S2S: speech-to-speech translation models. To use it, simply record your audio, or click the example to load. Read more at the links below."

article = "<p style='text-align: center'><a href='https://arxiv.org/abs/2107.05604' target='_blank'>Direct speech-to-speech translation with discrete units</a> | <a href='https://github.com/facebookresearch/fairseq/tree/main/examples/speech_to_speech' target='_blank'>Github Repo</a></p>"

examples = [
  ["enhanced_direct_s2st_units_audios_es-en_set2_source_12478_cv.flac","xm_transformer_s2ut_800m-es-en-st-asr-bt_h1_2022"],
]

io1 = gr.Interface.load("huggingface/facebook/xm_transformer_s2ut_800m-es-en-st-asr-bt_h1_2022", api_key=st.secrets["api_key"])
   
def inference(audio, model):
#    if mic is not None and file is None:
#        audio = mic
#    elif file is not None and mic is None:
#        audio = file
#    else:
#        return "ERROR: You must and may only select one method, it cannot be empty or select both methods at once."
    out_audio = io1(audio)   
    return out_audio 


gr.Interface(
    inference,
    [gr.inputs.Audio(source="microphone", type="filepath", label="Input"),gr.inputs.Dropdown(choices=["xm_transformer_s2ut_800m-es-en-st-asr-bt_h1_2022"], default="xm_transformer_s2ut_800m-es-en-st-asr-bt_h1_2022",type="value", label="Model")
],
    gr.outputs.Audio(label="Output"),
    article=article,
    title=title,
    description=description).queue().launch()