Update Documentation index, remove espeak-ng licenses

[Alex-Wengg]

Summary

• Add 12 missing entries to Documentation/README.md (Nemotron, Qwen3 ASR, TDT-CTC 110M, CTC Decoder Guide, Directory Structure, Choosing an API, benchmarks, voice quality comparison, model conversion, AMI subset benchmark)
• Remove unused Sources/FluidAudio/Frameworks/LICENSES/espeak-ng/ folder (4 license files, espeak-ng is no longer vendored)

Test plan

• Verify all new links in Documentation/README.md resolve to existing files
• Confirm no code references espeak-ng licenses

---

[claude]

Claude finished @Alex-Wengg's task —— View job

---

PR Review: Documentation index update and espeak-ng cleanup ✅

I've completed a comprehensive review of this PR that adds missing documentation entries and removes unused espeak-ng license files.

Todo List:

• Create pending GitHub review
• Get PR diff for detailed analysis
• Review Documentation/README.md changes
• Verify espeak-ng license removal is safe
• Check for any remaining espeak-ng references in codebase
• Verify all new documentation links exist
• Add inline review comments for any issues
• Submit comprehensive review

Review Summary

🔴 Found 1 blocking issue: Missing documentation file ASR/ChoosingAnAPI.md referenced in the index.

✅ Confirmed safe: espeak-ng license removal is completely safe - no code dependencies found.

✅ Verified: All other 9 new documentation links point to existing files.

The review has been submitted with detailed findings and actionable feedback.

[claude]

Code Review Summary

I've reviewed this PR and found one issue that needs to be addressed:

🔴 Blocking Issue

Missing Documentation File: The file Documentation/ASR/ChoosingAnAPI.md referenced on line 10 of Documentation/README.md doesn't exist in the repository. This will create a broken link in the documentation.

Resolution: Either create the missing file or remove the reference from the documentation index.

✅ Positive Findings

1. Safe License Removal: ✅ Confirmed that removing the Sources/FluidAudio/Frameworks/LICENSES/espeak-ng/ directory is safe. I searched the codebase and found no references to espeak-ng that would be broken by this removal.

2. Valid Documentation Additions: ✅ All other newly added documentation links point to existing files:

   • ASR/DirectoryStructure.md ✅
   • ASR/TDT-CTC-110M.md ✅
   • ASR/Nemotron.md ✅
   • ASR/Qwen3-ASR.md ✅
   • CtcDecoderGuide.md ✅
   • ASR/benchmarks100.md ✅
   • Diarization/BenchmarkAMISubset.md ✅
   • TTS/voice-quality.md ✅
   • ModelConversion.md ✅

3. Proper Documentation Hygiene: ✅ The cleanup removes unused license files (951 deletions) and adds comprehensive documentation links, improving overall project organization.

Test Plan Verification

• ✅ espeak-ng references: Confirmed no code references exist that would be broken by license removal
• ❌ Documentation links: Found 1 broken link (ASR/ChoosingAnAPI.md missing)

Please fix the missing documentation file before merging.

[github-actions]

PocketTTS Smoke Test ✅

Check	Result
Build	✅
Model download	✅
Model load	✅
Synthesis pipeline	✅
Output WAV	✅ (202.5 KB)

_{Runtime: 0m50s}

_{Note: PocketTTS uses CoreML MLState (macOS 15) KV cache + Mimi streaming state. CI VM lacks physical GPU — audio quality may differ from Apple Silicon.}

[github-actions]

Speaker Diarization Benchmark Results

Speaker Diarization Performance

Evaluating "who spoke when" detection accuracy

Metric	Value	Target	Status	Description
DER	15.1%	<30%	✅	Diarization Error Rate (lower is better)
JER	24.9%	<25%	✅	Jaccard Error Rate
RTFx	29.33x	>1.0x	✅	Real-Time Factor (higher is faster)

Diarization Pipeline Timing Breakdown

Time spent in each stage of speaker diarization

Stage	Time (s)	%	Description
Model Download	8.398	23.5	Fetching diarization models
Model Compile	3.599	10.1	CoreML compilation
Audio Load	0.060	0.2	Loading audio file
Segmentation	10.731	30.0	Detecting speech regions
Embedding	17.885	50.0	Extracting speaker voices
Clustering	7.154	20.0	Grouping same speakers
Total	35.778	100	Full pipeline

Speaker Diarization Research Comparison

Research baselines typically achieve 18-30% DER on standard datasets

Method	DER	Notes
FluidAudio	15.1%	On-device CoreML
Research baseline	18-30%	Standard dataset performance

Note: RTFx shown above is from GitHub Actions runner. On Apple Silicon with ANE:

• M2 MacBook Air (2022): Runs at 150 RTFx real-time
• Performance scales with Apple Neural Engine capabilities

_{🎯 Speaker Diarization Test • AMI Corpus ES2004a • 1049.0s meeting audio • 35.8s diarization time • Test runtime: 1m 41s • 03/29/2026, 12:27 AM EST}

[github-actions]

Offline VBx Pipeline Results

Speaker Diarization Performance (VBx Batch Mode)

Optimal clustering with Hungarian algorithm for maximum accuracy

Metric	Value	Target	Status	Description
DER	14.5%	<20%	✅	Diarization Error Rate (lower is better)
RTFx	4.75x	>1.0x	✅	Real-Time Factor (higher is faster)

Offline VBx Pipeline Timing Breakdown

Time spent in each stage of batch diarization

Stage	Time (s)	%	Description
Model Download	16.366	7.4	Fetching diarization models
Model Compile	7.014	3.2	CoreML compilation
Audio Load	0.097	0.0	Loading audio file
Segmentation	25.329	11.5	VAD + speech detection
Embedding	219.935	99.6	Speaker embedding extraction
Clustering (VBx)	0.700	0.3	Hungarian algorithm + VBx clustering
Total	220.785	100	Full VBx pipeline

Speaker Diarization Research Comparison

Offline VBx achieves competitive accuracy with batch processing

Method	DER	Mode	Description
FluidAudio (Offline)	14.5%	VBx Batch	On-device CoreML with optimal clustering
FluidAudio (Streaming)	17.7%	Chunk-based	First-occurrence speaker mapping
Research baseline	18-30%	Various	Standard dataset performance

Pipeline Details:

• Mode: Offline VBx with Hungarian algorithm for optimal speaker-to-cluster assignment
• Segmentation: VAD-based voice activity detection
• Embeddings: WeSpeaker-compatible speaker embeddings
• Clustering: PowerSet with VBx refinement
• Accuracy: Higher than streaming due to optimal post-hoc mapping

_{🎯 Offline VBx Test • AMI Corpus ES2004a • 1049.0s meeting audio • 246.0s processing • Test runtime: 4m 12s • 03/29/2026, 12:22 AM EST}

[github-actions]

Parakeet EOU Benchmark Results ✅

Status: Benchmark passed
Chunk Size: 320ms
Files Tested: 100/100

Performance Metrics

Metric	Value	Description
WER (Avg)	7.03%	Average Word Error Rate
WER (Med)	4.17%	Median Word Error Rate
RTFx	8.70x	Real-time factor (higher = faster)
Total Audio	470.6s	Total audio duration processed
Total Time	57.2s	Total processing time

Streaming Metrics

Metric	Value	Description
Avg Chunk Time	0.057s	Average chunk processing time
Max Chunk Time	0.114s	Maximum chunk processing time
EOU Detections	0	Total End-of-Utterance detections

_{Test runtime: 1m3s • 03/29/2026, 12:18 AM EST}

_{RTFx = Real-Time Factor (higher is better) • Processing includes: Model inference, audio preprocessing, state management, and file I/O}

[github-actions]

Qwen3-ASR int8 Smoke Test ✅

Check	Result
Build	✅
Model download	✅
Model load	✅
Transcription pipeline	✅
Decoder size	571 MB (vs 1.1 GB f32)

Performance Metrics

Metric	CI Value	Expected on Apple Silicon
Median RTFx	0.06x	~2.5x
Overall RTFx	0.06x	~2.5x

_{Runtime: 3m23s}

_{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.}

[github-actions]

VAD Benchmark Results

Performance Comparison

Dataset	Accuracy	Precision	Recall	F1-Score	RTFx	Files
MUSAN	92.0%	86.2%	100.0%	92.6%	726.0x faster	50
VOiCES	92.0%	86.2%	100.0%	92.6%	748.5x faster	50

Dataset Details

• MUSAN: Music, Speech, and Noise dataset - standard VAD evaluation
• VOiCES: Voices Obscured in Complex Environmental Settings - tests robustness in real-world conditions

✅: Average F1-Score above 70%

[github-actions]

ASR Benchmark Results ✅

Status: All benchmarks passed

Parakeet v3 (multilingual)

Dataset	WER Avg	WER Med	RTFx	Status
test-clean	0.57%	0.00%	4.74x	✅
test-other	1.35%	0.00%	3.21x	✅

Parakeet v2 (English-optimized)

Dataset	WER Avg	WER Med	RTFx	Status
test-clean	0.80%	0.00%	4.54x	✅
test-other	1.16%	0.00%	3.06x	✅

Streaming (v3)

Metric	Value	Description
WER	0.00%	Word Error Rate in streaming mode
RTFx	0.56x	Streaming real-time factor
Avg Chunk Time	1.633s	Average time to process each chunk
Max Chunk Time	1.816s	Maximum chunk processing time
First Token	1.926s	Latency to first transcription token
Total Chunks	31	Number of chunks processed

Streaming (v2)

Metric	Value	Description
WER	0.00%	Word Error Rate in streaming mode
RTFx	0.55x	Streaming real-time factor
Avg Chunk Time	1.650s	Average time to process each chunk
Max Chunk Time	2.181s	Maximum chunk processing time
First Token	1.733s	Latency to first transcription token
Total Chunks	31	Number of chunks processed

_{Streaming tests use 5 files with 0.5s chunks to simulate real-time audio streaming}

_{25 files per dataset • Test runtime: 6m14s • 03/29/2026, 12:32 AM EST}

_{RTFx = Real-Time Factor (higher is better) • Calculated as: Total audio duration ÷ Total processing time
Processing time includes: Model inference on Apple Neural Engine, audio preprocessing, state resets between files, token-to-text conversion, and file I/O
Example: RTFx of 2.0x means 10 seconds of audio processed in 5 seconds (2x faster than real-time)}

Expected RTFx Performance on Physical M1 Hardware:

• M1 Mac: ~28x (clean), ~25x (other)
• CI shows ~0.5-3x due to virtualization limitations

_{Testing methodology follows HuggingFace Open ASR Leaderboard}

[github-actions]

Sortformer High-Latency Benchmark Results

ES2004a Performance (30.4s latency config)

Metric	Value	Target	Status
DER	33.4%	<35%	✅
Miss Rate	24.4%	-	-
False Alarm	0.2%	-	-
Speaker Error	8.8%	-	-
RTFx	12.6x	>1.0x	✅
Speakers	4/4	-	-

_{Sortformer High-Latency • ES2004a • Runtime: 2m 38s • 2026-03-29T04:32:23.950Z}