docs: Add model documentation references to README highlights#501
docs: Add model documentation references to README highlights#501Alex-Wengg merged 6 commits intomainfrom
Conversation
Add links from README highlights section to Documentation/Models.md for: - ASR models (Parakeet TDT v3, Parakeet EOU) - TTS models (Kokoro, PocketTTS) - Diarization models - VAD models (Silero) Makes it easier for readers to find detailed model information.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
Keep only ASR model links and diarization section link. Remove individual model links for Kokoro, PocketTTS, and Silero VAD.
Keep only ASR model links in highlights section.
Add brief mention that ASR supports Japanese and Chinese in addition to the 25 European languages.
Changed from specifically mentioning 'Parakeet TDT v3' to 'Batch transcription with TDT and CTC models' to accurately represent the mix of model architectures available (TDT v2/v3, TDT-CTC-110M, CTC Japanese, TDT Japanese, CTC Chinese).
Lead with Parakeet TDT v3 (0.6b) as the main selling point while also mentioning other TDT/CTC models available for batch transcription.
Speaker Diarization Benchmark ResultsSpeaker Diarization PerformanceEvaluating "who spoke when" detection accuracy
Diarization Pipeline Timing BreakdownTime spent in each stage of speaker diarization
Speaker Diarization Research ComparisonResearch baselines typically achieve 18-30% DER on standard datasets
Note: RTFx shown above is from GitHub Actions runner. On Apple Silicon with ANE:
🎯 Speaker Diarization Test • AMI Corpus ES2004a • 1049.0s meeting audio • 38.7s diarization time • Test runtime: 2m 16s • 04/07/2026, 09:16 PM EST |
Parakeet EOU Benchmark Results ✅Status: Benchmark passed Performance Metrics
Streaming Metrics
Test runtime: 0m51s • 04/07/2026, 10:17 PM EST RTFx = Real-Time Factor (higher is better) • Processing includes: Model inference, audio preprocessing, state management, and file I/O |
VAD Benchmark ResultsPerformance Comparison
Dataset Details
✅: Average F1-Score above 70% |
PocketTTS Smoke Test ✅
Runtime: 0m40s Note: PocketTTS uses CoreML MLState (macOS 15) KV cache + Mimi streaming state. CI VM lacks physical GPU — audio quality may differ from Apple Silicon. |
Kokoro TTS Smoke Test ✅
Runtime: 0m39s Note: Kokoro TTS uses CoreML flow matching + Vocos vocoder. CI VM lacks physical ANE — performance may differ from Apple Silicon. |
Qwen3-ASR int8 Smoke Test ✅
Performance Metrics
Runtime: 5m9s Note: CI VM lacks physical GPU — CoreML MLState (macOS 15) KV cache produces degraded results on virtualized runners. On Apple Silicon: ~1.3% WER / 2.5x RTFx. |
ASR Benchmark Results ✅Status: All benchmarks passed Parakeet v3 (multilingual)
Parakeet v2 (English-optimized)
Streaming (v3)
Streaming (v2)
Streaming tests use 5 files with 0.5s chunks to simulate real-time audio streaming 25 files per dataset • Test runtime: 5m33s • 04/07/2026, 10:16 PM EST RTFx = Real-Time Factor (higher is better) • Calculated as: Total audio duration ÷ Total processing time Expected RTFx Performance on Physical M1 Hardware:• M1 Mac: ~28x (clean), ~25x (other) Testing methodology follows HuggingFace Open ASR Leaderboard |
Sortformer High-Latency Benchmark ResultsES2004a Performance (30.4s latency config)
Sortformer High-Latency • ES2004a • Runtime: 2m 34s • 2026-04-08T02:03:49.093Z |
Offline VBx Pipeline ResultsSpeaker Diarization Performance (VBx Batch Mode)Optimal clustering with Hungarian algorithm for maximum accuracy
Offline VBx Pipeline Timing BreakdownTime spent in each stage of batch diarization
Speaker Diarization Research ComparisonOffline VBx achieves competitive accuracy with batch processing
Pipeline Details:
🎯 Offline VBx Test • AMI Corpus ES2004a • 1049.0s meeting audio • 253.8s processing • Test runtime: 4m 17s • 04/07/2026, 10:25 PM EST |
1 participant
Summary
Adds links from README highlights section to ASR model documentation, highlighting Parakeet TDT v3 as the primary batch transcription model.
Changes
Updated the ASR highlight:
Before
Generic description without highlighting the flagship Parakeet TDT v3 model.
After
Readers see Parakeet TDT v3 as the primary batch transcription model and can click through to get full model details including:
Result
Better discoverability of ASR model documentation with Parakeet TDT v3 prominently featured as the main batch transcription model.