Skip to content

Clean up CI workflows: rename Kokoro, remove dead framework checks#463

Merged
Alex-Wengg merged 1 commit intomainfrom
chore/cleanup-workflows
Mar 29, 2026
Merged

Clean up CI workflows: rename Kokoro, remove dead framework checks#463
Alex-Wengg merged 1 commit intomainfrom
chore/cleanup-workflows

Conversation

@Alex-Wengg
Copy link
Copy Markdown
Member

@Alex-Wengg Alex-Wengg commented Mar 29, 2026

Summary

  • Rename tts-test.yml to kokoro-tts-test.yml and polish to match pocket-tts-test.yml style (dependency caching, PR result comments, 45min timeout, explicit Swift 6.1 setup, ffmpeg install, remove unused FLUIDAUDIO_ENABLE_TTS env var)
  • Delete framework-app-store-validation.yml and framework-validation.yml — both filter on Sources/FluidAudio/Frameworks/** which no longer exists, so they never trigger. framework-app-store-validation.yml also references a nonexistent FrameworkLinkTests test class.

Test plan

  • Verify Kokoro TTS workflow runs on this PR
  • Confirm no other workflows reference the deleted files

Open with Devin

- Rename tts-test.yml to kokoro-tts-test.yml and polish to match
  pocket-tts-test.yml (caching, PR comments, timeout, Swift 6.1,
  ffmpeg). Remove unused FLUIDAUDIO_ENABLE_TTS env var.
- Delete framework-app-store-validation.yml and framework-validation.yml
  which target Sources/FluidAudio/Frameworks/** (path no longer exists,
  workflows never trigger).
@claude
Copy link
Copy Markdown
Contributor

claude bot commented Mar 29, 2026

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


PR Review Complete ✅

Completed comprehensive review of PR #463: Clean up CI workflows: rename Kokoro, remove dead framework checks

Todo List:

  • Create pending PR review
  • Get PR diff to analyze changes and line numbers
  • Review new kokoro-tts-test.yml workflow for best practices
  • Verify rationale for deleting framework validation workflows
  • Add inline review comments for any issues found
  • Submit review as non-blocking comment
  • Provide final summary

Review Summary:

✅ Excellent cleanup with significant improvements:

  1. New kokoro-tts-test.yml - Major upgrade from old tts-test.yml:

    • Comprehensive caching strategy (models, Homebrew, build artifacts)
    • Professional PR automation with status tables and file size reporting
    • Proper error handling and 45-minute timeout protection
    • Modern GitHub Actions best practices
  2. Justified deletions - Both deleted workflows were dead code:

    • Filtered on Sources/FluidAudio/Frameworks/** (path doesn't exist)
    • framework-app-store-validation.yml referenced non-existent FrameworkLinkTests
    • No other files reference these workflows
  3. Verified test plan requirements:

    • ✅ Confirmed Sources/FluidAudio/Frameworks/ directory doesn't exist
    • ✅ No other workflows reference deleted files
    • ✅ Kokoro TTS workflow should run successfully on this PR

No blocking issues identified. This is a well-executed cleanup that removes dead code and significantly improves the TTS testing workflow.


Copy link
Copy Markdown
Contributor

@claude claude bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review Summary

This PR makes a focused and well-reasoned cleanup of CI workflows. The changes are solid overall with good improvements to the TTS workflow.

Excellent Improvements in kokoro-tts-test.yml:

  1. Comprehensive caching strategy (lines 25-34): Much better than original - includes model paths, Homebrew caches, and build artifacts
  2. Proper dependency management: Explicit Swift 6.1 setup and ffmpeg installation with fallback
  3. Enhanced error handling: Better smoke test logic with proper exit code capture and status reporting
  4. Professional PR automation: Excellent comment system with status table, file size reporting, and proper comment updating logic
  5. Security: Appropriate pull-requests: write permission and continue-on-error: true for non-blocking PR comments
  6. Timeout protection: 45-minute timeout prevents runaway jobs

Justified Deletions:

The deletion of framework-app-store-validation.yml and framework-validation.yml is completely justified:

  • Both workflows filter on Sources/FluidAudio/Frameworks/** which does not exist in the current codebase
  • framework-app-store-validation.yml references a non-existent FrameworkLinkTests test class
  • No other workflows or documentation reference these deleted files
  • These workflows would never trigger, making them dead code

🔧 Minor Observations:

  • The removal of FLUIDAUDIO_ENABLE_TTS: "1" environment variable appears intentional and likely unused
  • The workflow naming is now consistent with other test workflows in the repository
  • The new workflow follows modern GitHub Actions best practices

📋 Test Plan Verification:

  • ✅ Verified that Sources/FluidAudio/Frameworks/ directory does not exist
  • ✅ Confirmed no other workflows reference the deleted files
  • ✅ The Kokoro TTS workflow should run successfully on this PR

Overall Assessment: This is a high-quality cleanup that removes dead code and significantly improves the TTS testing workflow. No blocking issues identified.

Copy link
Copy Markdown
Contributor

@devin-ai-integration devin-ai-integration bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 3 additional findings.

Open in Devin Review

@github-actions
Copy link
Copy Markdown

PocketTTS Smoke Test ✅

Check Result
Build
Model download
Model load
Synthesis pipeline
Output WAV ✅ (176.3 KB)

Runtime: 0m29s

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
Copy link
Copy Markdown

Kokoro TTS Smoke Test ✅

Check Result
Build
Model download
Model load
Synthesis pipeline
Output WAV ✅ (634.8 KB)

Runtime: 0m47s

Note: Kokoro TTS uses CoreML flow matching + Vocos vocoder. CI VM lacks physical ANE — performance may differ from Apple Silicon.

@github-actions
Copy link
Copy Markdown

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 12.20x Real-time factor (higher = faster)
Total Audio 470.6s Total audio duration processed
Total Time 39.5s Total processing time

Streaming Metrics

Metric Value Description
Avg Chunk Time 0.040s Average chunk processing time
Max Chunk Time 0.079s Maximum chunk processing time
EOU Detections 0 Total End-of-Utterance detections

Test runtime: 1m8s • 03/29/2026, 01:38 AM EST

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

@Alex-Wengg Alex-Wengg merged commit 7eb11e2 into main Mar 29, 2026
16 checks passed
@Alex-Wengg Alex-Wengg deleted the chore/cleanup-workflows branch March 29, 2026 05:39
@github-actions
Copy link
Copy Markdown

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: 3m29s

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
Copy link
Copy Markdown

ASR Benchmark Results ✅

Status: All benchmarks passed

Parakeet v3 (multilingual)

Dataset WER Avg WER Med RTFx Status
test-clean 0.57% 0.00% 4.65x
test-other 1.75% 0.00% 3.13x

Parakeet v2 (English-optimized)

Dataset WER Avg WER Med RTFx Status
test-clean 0.80% 0.00% 4.60x
test-other 1.40% 0.00% 2.99x

Streaming (v3)

Metric Value Description
WER 0.00% Word Error Rate in streaming mode
RTFx 0.57x Streaming real-time factor
Avg Chunk Time 1.615s Average time to process each chunk
Max Chunk Time 1.792s Maximum chunk processing time
First Token 1.934s 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 0.53x Streaming real-time factor
Avg Chunk Time 1.687s Average time to process each chunk
Max Chunk Time 1.847s Maximum chunk processing time
First Token 1.648s 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: 6m58s • 03/29/2026, 01:44 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
Copy link
Copy Markdown

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 11.5x >1.0x
Speakers 4/4 - -

Sortformer High-Latency • ES2004a • Runtime: 3m 8s • 2026-03-29T05:45:16.186Z

@github-actions
Copy link
Copy Markdown

VAD Benchmark Results

Performance Comparison

Dataset Accuracy Precision Recall F1-Score RTFx Files
MUSAN 92.0% 86.2% 100.0% 92.6% 695.4x faster 50
VOiCES 92.0% 86.2% 100.0% 92.6% 707.4x 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
Copy link
Copy Markdown

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 5.28x >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 15.122 7.6 Fetching diarization models
Model Compile 6.481 3.3 CoreML compilation
Audio Load 0.055 0.0 Loading audio file
Segmentation 21.528 10.8 VAD + speech detection
Embedding 197.932 99.6 Speaker embedding extraction
Clustering (VBx) 0.729 0.4 Hungarian algorithm + VBx clustering
Total 198.823 100 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 • 220.2s processing • Test runtime: 3m 47s • 03/29/2026, 01:45 AM EST

@github-actions
Copy link
Copy Markdown

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 26.91x >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 10.717 27.5 Fetching diarization models
Model Compile 4.593 11.8 CoreML compilation
Audio Load 0.081 0.2 Loading audio file
Segmentation 11.694 30.0 Detecting speech regions
Embedding 19.490 50.0 Extracting speaker voices
Clustering 7.796 20.0 Grouping same speakers
Total 38.998 100 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 • 39.0s diarization time • Test runtime: 2m 17s • 03/29/2026, 01:51 AM EST

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant