Evaluation audio and objective scores for the Fasih-TTS-V1 Arabic (MSA / Fusha) text-to-speech model. Every clip is the model's own output, paired with its reference text, its Whisper-large-v3 transcription, and per-clip WER / CER.

Split (`test_set`)	Clips	Purpose	Mean CER
`silma_msa`	10	SILMA open-source Arabic TTS benchmark (MSA sentences)	2.0%
`samples`	3	General showcase (greeting, fiqh, reflection)	0.6%
`consistency`	8	Varied domain sentences	1.0%
`variance`	4	Same sentence synthesized ×4 (run-to-run stability)	2.0%
`stress`	6	Hard cases: long text, numbers, lists, term-heavy	2.1%

31 clips, 24 kHz mono. metadata.csv columns: file_name, text, test_set, wer_pct, cer_pct, asr_transcription.

SILMA benchmark result

On the SILMA Open-Source Arabic TTS Benchmark (MSA), Fasih-TTS-V1 ranks #1 by intelligibility (lowest WER/CER via Whisper-large-v3):

Rank	Model	WER%	CER%
1	Fasih-TTS-V1	6.5	2.0
2	XTTS (base)	10.3	6.4
3	silma_tts	11.1	6.0
4	chatterbox	12.8	6.6
5	omnivoice	15.3	7.2
6	habibi_specialized	21.9	9.7

Full 6-model per-clip provenance: silma_all_models_detailed.csv. WER/CER measure intelligibility, not naturalness — SILMA's own benchmark is a human auditory comparison.

Load

from datasets import load_dataset
ds = load_dataset("NightPrince/Fasih-TTS-Benchmark")
print(ds["train"][0])  # {'audio': ..., 'text': ..., 'test_set': ..., 'cer_pct': ...}

Methodology & reproduction

Audio synthesized with Fasih's production front-end (auto-diacritization via CATT + number expansion + chunking); scored with Whisper-large-v3 against normalized references (diacritics-stripped, digits expanded). Reproduce with scripts/silma_benchmark.py, scripts/silma_compare.py, scripts/build_eval_dataset.py in the model's GitHub repo.

License & copyright

Audio is output of Fasih-TTS-V1, a derivative of Coqui XTTS v2 (Coqui Public Model License — non-commercial). Released under CC-BY-NC-4.0. Copyright 2026 Yahya Elnawasany (NightPrince) — https://nightprincey.github.io/Portfolio-App/