Update app.py

app.py

@@ -1,42 +1,62 @@
 import gradio as gr
-import tempfile
 import numpy as np
+import soundfile as sf
+import tempfile
 from faster_whisper import WhisperModel
-from scipy.io.wavfile import write
 
+# Fastest practical model
 model = WhisperModel(
-    "tiny",
+    "turbo",
     device="cpu",
     compute_type="int8"
 )
 
-last_text = ""
+full_transcript = ""
+last_segment = ""
 
 def transcribe(audio):
-    global last_text
+
+    global full_transcript
+    global last_segment
 
     if audio is None:
-        return ""
+        return full_transcript
 
-    sr, y = audio
+    sr, data = audio
 
-    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
-        write(f.name, sr, y.astype(np.int16))
+    if len(data) < sr // 2:
+        return full_transcript
 
-        segments, _ = model.transcribe(
+    with tempfile.NamedTemporaryFile(suffix=".wav") as f:
+
+        sf.write(f.name, data, sr)
+
+        segments, info = model.transcribe(
             f.name,
-            language="en"
+            language="en",
+            vad_filter=True,
+            beam_size=1,
+            best_of=1,
+            temperature=0
         )
 
-    text = " ".join(
-        segment.text for segment in segments
-    ).strip()
+        current_text = " ".join(
+            segment.text.strip()
+            for segment in segments
+        )
 
-    if text == last_text:
-        return last_text
+    if not current_text:
+        return full_transcript
 
-    last_text = text
-    return text
+    if current_text != last_segment:
+
+        if full_transcript:
+            full_transcript += " "
+
+        full_transcript += current_text
+        last_segment = current_text
+
+    return full_transcript
 
 
 with gr.Blocks() as demo:
@@ -44,20 +64,21 @@ with gr.Blocks() as demo:
     gr.Markdown("# Real-Time English Speech Recognition")
 
     audio = gr.Audio(
-        streaming=True,
         sources=["microphone"],
+        streaming=True,
         type="numpy"
     )
 
-    text = gr.Textbox(
-        label="Live Transcript",
-        lines=10
+    output = gr.Textbox(
+        label="Transcript",
+        lines=12
     )
 
     audio.stream(
-        transcribe,
-        audio,
-        text
+        fn=transcribe,
+        inputs=audio,
+        outputs=output,
+        stream_every=1
     )
 
 demo.launch()