Update app.py

app.py

@@ -1,49 +1,89 @@
 from transformers import pipeline
+import torch
 import gradio as gr
-from pyctcdecode import BeamSearchDecoderCTC
+import subprocess
+import numpy as np
+import time
 
-#lmID = "aware-ai/german-lowercase-wiki-5gram"
-#decoder = BeamSearchDecoderCTC.load_from_hf_hub(lmID)
-p = pipeline("automatic-speech-recognition", model="aware-ai/wav2vec2-xls-r-1b-5gram-german")
-ttp = pipeline("text2text-generation", model="aware-ai/marian-german-grammar")
+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', force_reload=False, onnx=True)
+                              
+def ffmpeg_read(bpayload: bytes, sampling_rate: int) -> np.array:
+    """
+    Helper function to read an audio file through ffmpeg.
+    """
+    ar = f"{sampling_rate}"
+    ac = "1"
+    format_for_conversion = "f32le"
+    ffmpeg_command = [
+        "ffmpeg",
+        "-i",
+        "pipe:0",
+        "-ac",
+        ac,
+        "-ar",
+        ar,
+        "-f",
+        format_for_conversion,
+        "-hide_banner",
+        "-loglevel",
+        "quiet",
+        "pipe:1",
+    ]
 
-def transcribe(audio):
-    transcribed = p(audio, chunk_length_s=16, stride_length_s=(4, 0))["text"]
-    return transcribed
-    
-def punctuate(transcribed):
-    punctuated = ttp(transcribed, max_length = 512)[0]["generated_text"]
-    return punctuated
-
-def get_asr_interface():
-    return gr.Interface(
-        fn=transcribe, 
-        inputs=[
-            gr.inputs.Audio(source="microphone", type="filepath")
-        ],
-        outputs=[
-            "textbox",
-        ])
+    try:
+        with subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE) as ffmpeg_process:
+            output_stream = ffmpeg_process.communicate(bpayload)
+    except FileNotFoundError as error:
+        raise ValueError("ffmpeg was not found but is required to load audio files from filename") from error
+    out_bytes = output_stream[0]
+    audio = np.frombuffer(out_bytes, np.float32)
+    if audio.shape[0] == 0:
+        raise ValueError("Malformed soundfile")
+    return audio
+
+(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": ""}
+    with open(audio, "rb") as f:
+            payload = f.read()
+    audio = ffmpeg_read(payload, sampling_rate=16000)   
+    _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
         
-def get_punctuation_interface():
-    return gr.Interface(
-        fn=punctuate, 
-        inputs=[
-            "textbox",
-        ],
-        outputs=[
-            "textbox",
-        ])
-
-interfaces = [
-    get_asr_interface(),
-    get_punctuation_interface(),
-]
-
-names = [
-    "ASR",
-    "GRAMMAR",
-]
-
-gr.TabbedInterface(interfaces, names).launch(server_name = "0.0.0.0", enable_queue=False)
+        state["text"] += state["temp_text"]
+        state["temp_text"] = ""
+        state["audio"] = ""
+
+    time.sleep(0.5)
+    return f'{state["text"]} ( {state["temp_text"]} )', state
+
+gr.Interface(
+    transcribe, 
+    [gr.Audio(source="microphone", type="filepath", streaming=True), "state"],
+    
+    [gr.Textbox(),"state"],
+    live=True
+    ).launch(server_name = "0.0.0.0")