🤗 Diarizers Community — Open-Source ASR, Speaker Diarization, and DiarizationLM Post-Processing

diarizers-community builds open-source models and software for speaker-attributed automatic speech recognition (ASR + Speaker Diarization + DiarizationLM Post-Processing):

• Unified Python Package & CLI (diarizers v1.0.0):
  • GitHub: https://github.com/diarizers-community/diarizers
  • PyPI: https://pypi.org/project/diarizers/ (pip install diarizers)
• Pretrained Multi-Domain DiarizationLM Model (Gemma 4 E4B, Safetensors + Q4_K_M GGUF):
  • Model Hub: diarizers-community/DiarizationLM-Gemma-4-E4B-v1
• Interactive Demos (Hugging Face Spaces):
  • End-to-End Audio Pipeline Demo (diarizers): spaces/diarizers-community/diarizers
  • Text-Only DiarizationLM GGUF Demo: spaces/diarizers-community/DiarizationLM-GGUF

Architecture of diarizers (v1.0.0)

diarizers (v1.0.0) provides a 100% Apache-2.0 / MIT 4-stage pipeline for speaker-attributed transcription and LLM post-processing:

1. Stage 1 — Timestamped ASR (diarizers.asr): openai/whisper-large-v3-turbo (WhisperWordASR, MIT License) with native Whisper <|t.tt|> segment-level timestamps and character-length-proportional word timestamp interpolation (plus support for Qwen/Qwen3-ASR GGUF via audio.cpp, Apache-2.0 License).
2. Stage 2 — Neural Speaker Turn Segmentation + Speaker Embeddings + Spectral Clustering (diarizers.diarization & diarizers.spectral):
   • Standalone PyTorch pyannote/segmentation-3.0 (PyanNetSegmentation: SincNet + 4-layer BiLSTM + 7-class powerset classifier, MIT License) for local speaker-change flux boundary snapping.
   • Standalone PyTorch speechbrain/spkrec-ecapa-voxceleb (SpeechBrainECAPATDNN, Apache-2.0 License) with exact HTK decibel-scaled (10 * log10) 80-bin log-mel filterbanks, global embedding mean subtraction, and recording-level channel centering.
   • Clean-room Symmetric Normalized Laplacian Spectral Clusterer (L_sym = I - D^{-1/2} W D^{-1/2}, Apache-2.0 License) with Gaussian temporal smoothing and calibrated eigengap speaker count estimation (eigen_threshold=0.45).
3. Stage 3 — Word-to-Turn Alignment & Chunked Prompting (diarizers.alignment): Sub-millisecond assign_speakers overlap orchestration, nearest-turn midpoint fallback, and 4,000-character chunked prompt generation (ChunkedPromptBuilder).
4. Stage 4 — DiarizationLM-Gemma-4-E4B-v1 Post-Processing & LG-TPST (diarizers.llm & diarizers.alignment): Post-processing via diarizers-community/DiarizationLM-Gemma-4-E4B-v1 (CPU llama.cpp Q4_K_M GGUF or GPU/CPU Hugging Face transformers), Hungarian inter-chunk permutation reconciliation, Myers' (1999) bit-parallel exact word-level Levenshtein alignment (tpst_align), and Locality-Gated Transcript-Preserving Speaker Transfer (LG-TPST).

Benchmark Evaluation Results

All metrics are micro-averaged across the full evaluation splits (172 Fisher conversations, 20 CALLHOME-English calls, 3 ICSI meetings, and 16 AMI Mix-Headset meetings = 211 test recordings).

1. Open-Source Frontend (diarizers v1.0.0) + DiarizationLM-Gemma-4-E4B-v1

Note: Post-LLM evaluation numbers above were computed with the full-precision (bfloat16) DiarizationLM-Gemma-4-E4B-v1 checkpoint using 6,000-character prompt chunks. For CPU deployment, the repository also provides the 4-bit quantized DiarizationLM-Gemma-4-E4B-v1-q4_k_m.gguf model (4,000-character default chunks).

2. USM + Turn-to-Diarize Frontend + DiarizationLM-Gemma-4-E4B-v1

Quickstart

pip install "diarizers[audio,llm]"

import diarizers

pipe = diarizers.DiarizersPipeline.from_pretrained(
    asr_model="openai/whisper-large-v3-turbo",
    diarizer_preset=diarizers.FISHER_TELEPHONY_PRESET,
    llm_model_or_path="diarizers-community/DiarizationLM-Gemma-4-E4B-v1",
    llm_gguf_filename="DiarizationLM-Gemma-4-E4B-v1-q4_k_m.gguf",
    llm_backend="gguf",
    device="cpu",
)

transcript = pipe("conversation.wav", num_speakers=2)
print(transcript.to_diarized_text())
print(transcript.to_rttm(recording_id="conversation"))
print(transcript.to_srt())

For full documentation, CLI usage, SFT training recipes, and evaluation tools, visit https://github.com/diarizers-community/diarizers.