first test of nb small

app.py

@@ -0,0 +1,54 @@
+import gradio as gr
+from transformers import pipeline
+import numpy as np
+
+model_id = "NbAiLab/nb-whisper-small-beta"
+transcriber = pipeline("automatic-speech-recognition", model=model_id)
+total_time = 0
+counter = 0
+
+
+def make_timestamp(ref):
+    global total_time
+    hh = int((total_time + ref) / 3600)
+    mm = int((total_time + ref) / 60) % 60
+    ss = int((total_time + ref) % 60)
+    mmm = int((total_time + ref) % 1000)
+    return f"{hh:02d}:{mm:02d}:{ss:02d},{mmm:03d}"
+
+
+def transcribe(audio):
+    global counter
+    global total_time
+    sr, y = audio
+    y = y.astype(np.float32)
+    y /= np.max(np.abs(y))
+
+    conf = {"sampling_rate": sr, "raw": y}
+    kwargs = {"task": "transcribe", "language": "no"}
+    res = transcriber(conf, generate_kwargs=kwargs, return_timestamps=True)
+    chunks = res["chunks"]
+    timestamps = [c["timestamp"] for c in chunks]
+    text = [c["text"].strip() for c in chunks]
+
+    entries = []
+    for (start, end), txt in zip(timestamps, text):
+        start_srt = make_timestamp(start)
+        end_srt = make_timestamp(end)
+        srt_entry = f"{counter}\n{start_srt} --> {end_srt}\n{txt}\n"
+        entries.append(srt_entry)
+        total_time += end
+        counter += 1
+
+    return "\n".join(entries)
+   
+
+
+demo = gr.Interface(
+    transcribe,
+    gr.Audio(source="microphone"),
+    "text",
+)
+
+demo.launch()
+