Add Speech Diarization + ASR (asr_diarize)

[wannaphong]

Add Speech Diarization + ASR (asr_diarize) by Sherpa-ONNX Diarization

Functions

Speech Diarization (Who Spoke When)

Detect speaker turns and timestamps using ONNX:

from pythaiasr import diarize

# Identify who spoke when
segments = diarize("meeting.wav")
for seg in segments:
    print(f"[{seg['start']:.2f}s - {seg['end']:.2f}s] {seg['speaker']}")

Speech Diarization + ASR (asr_diarize)

Detect speakers and transcribe each speaker turn with ASR:

from pythaiasr import asr_diarize

# Attributed transcription per speaker
turns = asr_diarize("meeting.wav", asr_model="typhoon_asr")
for turn in turns:
    print(f"[{turn['start']:.2f}s - {turn['end']:.2f}s] {turn['speaker']}: {turn['text']}")

See examples/diarize_example.py

============================================================
1. Speech Diarization (Who Spoke When)
============================================================
Processing: examples/../tests/test-diarize.wav ...
[  0.17s ->   1.87s] SPEAKER_02
[  1.97s ->   4.30s] SPEAKER_01
[  4.83s ->   6.49s] SPEAKER_02
[  6.78s ->   8.36s] SPEAKER_01

============================================================
2. Diarization + Speech Recognition (ASR Diarize)
============================================================
Transcribing turns with Typhoon ASR: examples/../tests/test-diarize.wav ...
[  0.17s ->   1.87s] SPEAKER_02: สวัสดีชาวโลกทุกท่าน
[  1.97s ->   4.30s] SPEAKER_01: แล้วระบบนี้ทํางานอย่างไร
[  4.83s ->   6.49s] SPEAKER_02: ใช้ปัญญาประดิษฐ์ในการทดสอบ
[  6.78s ->   8.36s] SPEAKER_01: ใช้งานได้ดีทีเดียว