Create app.py

app.py

@@ -0,0 +1,53 @@
+from transformers import pipeline
+import torch
+import gradio as gr
+import librosa
+import numpy as np
+import os
+
+
+p = pipeline("automatic-speech-recognition", model="flozi00/wav2vec2-xls-r-1b-5gram-german", use_auth_token = os.environ['auth'])
+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": None}):
+    if state is None:
+        state={"text": "", "temp_text": "", "audio": None}
+    wav_data, _sr = librosa.load(audio, sr=16000)
+    speech = is_speech(wav_data, _sr)
+    if(speech):
+        if(state["audio"] is None):
+            state["audio"] = wav_data
+        else:
+            state["audio"] = np.concatenate((state["audio"], wav_data))
+
+        text = p(state["audio"])["text"] + "\n"
+        state["temp_text"] = text
+    else:
+        state["text"] += state["temp_text"]
+        state["temp_text"] = ""
+        state["audio"] = None
+
+    return f'{state["text"]} ( {state["temp_text"]} )', state
+
+gr.Interface(
+    fn=transcribe, 
+    inputs=[
+        gr.inputs.Audio(source="microphone", type="filepath"), 
+        "state"
+    ],
+    outputs=[
+        "textbox",
+        "state"
+    ],
+    live=True).launch()