Update app.py

app.py

@@ -1,51 +1,15 @@
 from transformers import pipeline
-import torch
 import gradio as gr
-import librosa
-import numpy as np
-import time
 
 p = pipeline("automatic-speech-recognition", model="aware-ai/wav2vec2-base-german")
 
-model, utils = torch.hub.load(repo_or_dir='snakers4/silero-vad',
-                              model='silero_vad')
-
-(get_speech_timestamps,
- _, read_audio,
- *_) = utils
-
-def is_speech(wav, sr):
-    speech_timestamps = get_speech_timestamps(wav, model,
-                                    sampling_rate=sr)
-
-    return len(speech_timestamps) > 0
-
-def transcribe(audio, state={"text": "", "temp_text": "", "audio": ""}):
-    if state is None:
-        state={"text": "", "temp_text": "", "audio": ""}
-    wav_data, _sr = librosa.load(audio, sr=16000)
-    speech = is_speech(wav_data, _sr)
-    if(speech):
-        if(state["audio"] is ""):
-            state["audio"] = wav_data
-        else:
-            state["audio"] = np.concatenate((state["audio"], wav_data))
-    else:
-        if(state["audio"] is not ""):
-            text = p(state["audio"])["text"] + "\n"
-            state["temp_text"] = text
-        
-        state["text"] += state["temp_text"]
-        state["temp_text"] = ""
-        state["audio"] = ""
-
-    time.sleep(0.5)
-    return f'{state["text"]} ( {state["temp_text"]} )', state
+def transcribe(audio):
+    return p(audio, chunk_length_s=10, stride_length_s=(4, 2))["text"]
 
 gr.Interface(
     transcribe, 
-    [gr.Audio(source="microphone", type="filepath", streaming=True), "state"],
+    [gr.Audio(source="microphone", type="filepath")],
     
-    [gr.Textbox(),"state"],
+    [gr.Textbox()],
     live=True
     ).launch(server_name = "0.0.0.0")