spum / models /TEST_RESULTS.json
vinayapathak's picture
Publish tested English speech model packs with licences and checksums
23fa293 verified
Raw History Blame Contribute Delete
1.53 kB
{
"date": "2026-11-02",
"hardware": "Apple M5 Max",
"runtime": "Spum native speech integration, CPU, 4 threads",
"recognition": {
"model": "Whisper base.en, int8 ONNX",
"recordings": 3,
"reference_words": 77,
"word_errors": 3,
"word_error_rate": 0.03896103896103896,
"normalization": "Lowercase alphanumeric word tokens; punctuation excluded. Spelling and word-boundary variants remain errors.",
"inference_seconds": [
0.234429292,
0.604309875,
0.164565292
],
"audio_seconds": [
6.625,
16.715,
4.825
]
},
"synthesis": {
"model": "Kokoro v1.0 int8",
"checks": [
{
"voice": "af_heart",
"audio_seconds": 5.409708333333334,
"sample_rate": 24000,
"inference_seconds": 2.497420667,
"roundtrip_word_errors": 0,
"reference_words": 12
},
{
"voice": "af_alloy",
"audio_seconds": 3.369125,
"sample_rate": 24000,
"inference_seconds": 1.776452083,
"roundtrip_word_errors": 0,
"reference_words": 9
},
{
"voice": "am_adam",
"audio_seconds": 2.7133333333333334,
"sample_rate": 24000,
"inference_seconds": 1.278524416,
"roundtrip_word_errors": 0,
"reference_words": 7
}
]
},
"scope": "Small functional checks, not a general recognition benchmark, listening study or clinical validation. TTS roundtrip recognition is not independent perceptual evaluation."
}