Benchmark on-device ASR, TTS, VAD, and diarization latency and efficiency.
npx skills add https://github.com/soniqo/speech-swift --skill benchmark-soniqo
Please help me install this Agent Skill. Skill: benchmark Source: https://github.com/soniqo/speech-swift/tree/main/.claude/skills/benchmark Command: npx skills add https://github.com/soniqo/speech-swift --skill benchmark-soniqo
Benchmarking on-device ASR, TTS, VAD, and diarization performance can be hard to compare across models, builds, and hardware, making it difficult to validate real-world latency and accuracy.
Run the ASR benchmark on the provided test audio and collect the resulting RTF by invoking the benchmark module as asr.
High-intent search queries and answers about installing and using this skill.
Benchmark speech recognition latency on Apple Silicon by executing provided Python scripts against a release build binary to quantify real-time factor and processing efficiency for ASR tasks.
Yes, you can measure text to speech and voice activity detection performance separately by invoking module-scoped benchmark arguments to run isolated performance tests for TTS and VAD without altering your workflow.
To benchmark diarization accuracy, you need the provided test audio files and VoxConverse test sets to ensure repeatable comparison of real-world latency and efficiency across different builds.
Yes, a release build binary is required to measure on-device speech processing performance because it ensures consistent performance measurement without the overhead of debug instrumentation.
Validate real-time factor targets by running the ASR benchmark module on provided test audio to collect resulting RTF metrics and compare them against your expected performance thresholds.
d-o-hubValidate Rust crates with cargo check, test, fmt, clippy, LOC, and benchmarks.
ad-repoAnalyze and optimize NullPlayer's audio playback pipeline for local and streaming sources.
Canhada-LabsIntegrate ASR and TTS providers with real-time audio streaming and speaker diarization.
xenitV1Automate Lighthouse-based performance profiling to identify web application bottlenecks.
