IzwiIzwi

izwi bench

izwi bench

Run performance benchmarks.

Synopsis

izwi bench <COMMAND>

Subcommands

CommandDescription
ttsBenchmark TTS inference
asrBenchmark ASR inference
throughputBenchmark system throughput

izwi bench tts

Benchmark text-to-speech performance.

izwi bench tts [OPTIONS]

Options

OptionDescriptionDefault
-m, --model <MODEL>Model to benchmarkqwen3-tts-0.6b-base
-i, --iterations <N>Number of iterations10
-t, --text <TEXT>Text to synthesizeDefault test text
--warmupEnable warmup iteration—

Example

izwi bench tts --model qwen3-tts-0.6b-base --iterations 20 --warmup

izwi bench asr

Benchmark speech recognition performance.

izwi bench asr [OPTIONS]

Options

OptionDescriptionDefault
-m, --model <MODEL>Model to benchmarkqwen3-asr-0.6b
-i, --iterations <N>Number of iterations10
-f, --file <PATH>Audio file to useBuilt-in test audio
--warmupEnable warmup iteration—

Example

izwi bench asr --model qwen3-asr-0.6b --file test.wav --iterations 20

izwi bench throughput

Benchmark overall system throughput.

izwi bench throughput [OPTIONS]

Options

OptionDescriptionDefault
-d, --duration <SECONDS>Test duration30
-c, --concurrent <N>Concurrent requests1

Example

izwi bench throughput --duration 60 --concurrent 4

Output

Benchmarks report:

  • Latency — Average, min, max, p50, p95, p99
  • Throughput — Requests per second
  • Tokens/second — For TTS benchmarks
  • Real-time factor — Audio duration vs processing time

See Also