app.py

import os
import gradio as gr
import torch
from transformers import pipeline


title = "Transcribe speech in several languages"
device = "cuda:0" if torch.cuda.is_available() else "cpu"

asr_pipe_audio2Text_Ge = pipeline(task="automatic-speech-recognition", model="jonatasgrosman/wav2vec2-large-xlsr-53-german")
asr_pipe_whisper = pipeline(task="automatic-speech-recognition", model="openai/whisper-medium", device=device)

def transcribeFile(inputlang, audio_path : str) -> str:
    #transcription = asr_pipe_audio2Text_Ge(audio_path)
    #transcription = asr_pipe_whisper(audio_path, max_new_tokens=256, generate_kwargs={"task":"transcribe"})
    if inputlang == "Auto Detect":
        transcription = asr_pipe_whisper(audio_path, chunk_length_s=10, stride_length_s=(4, 2), generate_kwargs={"task":"transcribe"}, batch_size=32)
    elif inputlang == "German":
        transcription = asr_pipe_audio2Text_Ge(audio_path, chunk_length_s=10, stride_length_s=(4, 2), batch_size=32)    
    return transcription["text"]

def translateAudio(audio_path):
    translationOutput = asr_pipe_whisper(audio_path, max_new_tokens=256, generate_kwargs={"task":"translate"})
    return translationOutput

def transcribeFileMulti(inputlang, audio_path : str) -> str:
    if inputlang == "English":
        transcription = asr_pipe_whisper(audio_path)
    elif inputlang == "German":
        transcription = asr_pipe_audio2Text_Ge(audio_path)
        translation = translateAudio(audio_path)
        t1 = transcription["text"]
        t2 = translation["text"]
        output = t1+t2
    return output #transcription["text"]


    
app1 = gr.Interface(
    fn=transcribeFile,
    #inputs=gr.inputs.Audio(label="Upload audio file", type="filepath"),
    inputs=[gr.Radio(["Auto Detect", "German"], value="Auto Detect", label="Source Language", info="Select the language of the speech you want to transcribe"),
                     gr.Audio(source="upload", type="filepath",label="Upload audio file")],     
    outputs="text",
    title=title
)


app2 = gr.Interface(
    fn=transcribeFileMulti,
    inputs=[gr.Radio(["Auto Detect", "German"], value="Auto Detect", label="Source Language", info="Select the language of the speech you want to transcribe"),
                     gr.Audio(source="microphone", type="filepath")], 
    outputs="text",
    title=title
)


demo = gr.TabbedInterface([app1, app2], ["Audio File", "Microphone"])

if __name__ == "__main__":
    demo.launch()