Diarizers Audio Pipeline

🎙️ Diarizers v1.0.1 — End-to-End Audio ASR, Speaker Diarization & DiarizationLM

End-to-end open-source (Apache-2.0 / MIT) pipeline combining Stage 1: Whisper ASR (WhisperWordASR), Stage 2: Neural Speaker Segmentation + Embeddings + Spectral Clustering (pyannote/segmentation-3.0 + speechbrain/spkrec-ecapa-voxceleb + Symmetric Normalized Laplacian SpectralClusterer), Stage 3: Word-to-Turn Alignment, and Stage 4: DiarizationLM-Gemma-4-E4B-v1 Post-Processing (Q4_K_M GGUF + Locality-Gated Transcript-Preserving Speaker Transfer LG-TPST).

1. End-to-End Audio Pipeline Configuration

1. Input Conversation Audio (.wav / .mp3 / .flac)

Upload any multi-speaker audio file, or click one of the built-in conversation examples below.

2. Stage 1 — Whisper ASR Model

3. Stage 2 — Acoustic Diarization Preset

4. Known Speaker Count (2; 0 = Auto-detect via Laplacian Eigengap)

Stage 4: Enable DiarizationLM-Gemma-4-E4B-v1 Boundary Refinement

Enable Locality-Gated Speaker Transfer (LG-TPST)

Diarized Transcripts, RTTM & SRT Subtitles

Loading built-in conversation examples...

Before DiarizationLM (Acoustic Baseline: Whisper + PyanNet + ECAPA-TDNN + SpectralClusterer)

After DiarizationLM-Gemma-4-E4B-v1 (Refined Transcript — relabeled boundary words highlighted in green)

NIST RTTM Output

SRT Subtitles

🎙️ Diarizers v1.0.1 — End-to-End Audio ASR, Speaker Diarization & DiarizationLM

1. End-to-End Audio Pipeline Configuration

Diarized Transcripts, RTTM & SRT Subtitles

Built-in Multi-Speaker Conversation Examples (Click a row to load & run)