lemura-arabic-asr-lite

Lemura Labs' Arabic-first speech-recognition model — leaderboard-grade at a fraction of the size.

Arabic accuracy, without the GPU bill.

Overview · Benchmarks · Efficiency · Inference · Live Demo · Sibling model

---

What it is

lemura-arabic-asr-lite is a compact multi-dialect Arabic ASR model built for accuracy and efficiency. It is a FastConformer-CTC acoustic model adapted in-house from the NVIDIA FastConformer foundation — the contribution is the adaptation (the Arabic data curriculum and dialect coverage), not the foundation.

• Small and fast — ~115M parameters; runs comfortably on CPU and in real time, no GPU required.
• Dialect-aware — fine-tuned on ~2,900 hours of Arabic spanning MSA and the Gulf, Egyptian, Levantine and Maghrebi dialect groups, not MSA-only.
• Robust on real audio — strongest on broadcast, conversational and Gulf/MSA speech.
• Open and simple — a single .nemo file, loadable in a few lines with NVIDIA NeMo.

The result is leaderboard-grade Arabic ASR at edge scale — #2 of 36 systems on the Open Universal Arabic ASR Leaderboard, ahead of every 2B–30B audio-LLM evaluated, at ~115M parameters.

Model summary

Model	lemura-arabic-asr-lite — compact multi-dialect Arabic ASR
Task	Automatic speech recognition (audio → text)
Approach	Discriminative ASR — FastConformer encoder + CTC decoder
Training	adapted from the NVIDIA FastConformer foundation; ~2,900 h Arabic fine-tuning across 5 dialect groups
Total parameters	~115,000,000 (~0.12B)
Checkpoint size	~0.4 GB (single .nemo file)
Audio input	16 kHz mono (auto-resampled)
Languages	Arabic — MSA + Gulf / Egyptian / Levantine / Maghrebi
Runtime	NVIDIA NeMo — CPU · GPU · real-time
License	CC-BY-4.0

Benchmarks

Arabic dialectal ASR is hard — heavily dialectal, conversational, code-switched speech is the frontier for every system. On the Open Universal Arabic ASR Leaderboard, lemura-arabic-asr-lite ranks #2 of 36 systems with an average WER of 25.08 % — ahead of every model evaluated except one, including systems 17× to 250× larger.

Open Universal Arabic ASR Leaderboard — full standings

Per-dataset WER % across all six leaderboard test sets. Lower is better; Avg WER is the ranking metric. Our rows were produced with the official leaderboard code (same normalizer, same WER metric). Ours in bold.

#	Model	Params	Avg WER	SADA	CV-18	MASC-clean	MASC-noisy	MGB-2	Casablanca
1	lemuralabs/lemura-arabic-asr-lite (Ours)	0.12B	25.08	37.28	9.74	7.27	23.65	14.33	58.24
2	CohereLabs/cohere-transcribe-arabic-07-2026	~2B	25.87	37.47	5.82	19.60	27.07	15.54	49.71
3	omnilingual-asr/omniASR_LLM_7B	7B	28.32	41.61	8.75	19.69	29.29	14.13	56.46
4	omnilingual-asr/omniASR_LLM_3B	3B	29.96	46.18	9.15	19.90	30.03	14.22	60.27
5	omnilingual-asr/omniASR_LLM_1B	1B	29.96	43.84	9.55	20.03	30.26	15.34	60.68
6	CohereLabs/cohere-transcribe-03-2026	~2B	30.67	60.11	8.17	8.66	19.01	25.33	62.71
7	Qwen/Qwen3-Omni-30B-A3B-Instruct	30B	30.71	44.82	11.46	21.47	30.85	13.09	62.55
8	nvidia-conformer-ctc-large-arabic (lm)	0.6B	32.91	44.52	8.80	23.74	34.29	17.20	68.90
9	omnilingual-asr/omniASR_LLM_300M	0.3B	32.96	51.38	12.03	20.66	32.45	16.58	64.64
10	google/gemma-4-E4B-it	4B	32.98	43.40	19.65	24.86	33.59	17.72	58.63
11	Qwen/Qwen3-ASR-1.7B	1.7B	33.36	45.53	16.90	24.37	34.29	16.57	64.47
12	mistralai/Voxtral-Small-24B-2507	24B	34.47	50.82	15.25	23.96	34.43	16.03	66.30
13	nvidia-conformer-ctc-large-arabic (greedy)	0.6B	34.74	47.26	10.60	24.12	35.64	19.69	71.13
14	google/gemma-4-E2B-it	2B	35.87	46.23	23.76	27.47	36.15	20.72	60.87
15	openai/whisper-large-v3	1.5B	36.86	55.96	17.83	24.66	34.63	16.26	71.81
16	omnilingual-asr/omniASR_CTC_3B	3B	37.78	69.85	14.19	21.48	34.60	18.96	67.58
17	omnilingual-asr/omniASR_CTC_7B	7B	38.12	72.69	12.47	21.08	35.04	20.43	67.02
18	facebook/seamless-m4t-v2-large	2.3B	38.16	62.52	21.70	25.04	33.24	20.23	66.25
19	omnilingual-asr/omniASR_CTC_1B	1B	39.29	71.42	17.55	22.76	35.73	19.96	68.32
20	openai/whisper-large-v3-turbo	0.8B	40.05	60.36	25.73	25.51	37.16	17.75	73.79
21	openai/whisper-large-v2	1.5B	40.20	57.46	21.77	27.25	38.55	25.17	71.01
22	Qwen/Qwen3-ASR-0.6B	0.6B	42.19	53.75	28.28	31.34	42.63	25.45	71.68
23	openai/whisper-large	1.5B	42.57	63.24	26.04	28.89	40.79	24.28	72.18
24	mistralai/Voxtral-Mini-3B-2507	3B	42.58	63.65	22.12	28.37	41.27	22.56	77.52
25	asafaya/hubert-large-arabic-transcribe	0.3B	45.50	67.82	8.01	32.94	50.16	37.51	76.53
26	openai/whisper-medium	0.8B	45.57	67.71	28.07	29.99	42.91	29.32	75.44
27	nvidia-Parakeet-ctc-1.1b-concat	1.1B	46.54	70.70	26.34	30.49	45.95	24.94	80.80
28	omnilingual-asr/omniASR_CTC_300M	0.3B	46.65	78.11	27.90	28.40	43.26	26.85	75.35
29	nvidia-Parakeet-ctc-1.1b-universal	1.1B	51.96	73.58	40.01	36.16	50.03	30.68	81.30
30	microsoft/VibeVoice-ASR	—	52.99	69.83	44.25	32.95	52.43	25.10	93.37
31	facebook/mms-1b-all	1B	54.54	77.48	26.52	38.82	57.33	39.16	87.95
32	openai/whisper-small	0.24B	55.13	78.02	24.18	35.93	56.36	48.64	87.64
33	whitefox123/w2v-bert-2.0-arabic-4	0.6B	58.13	87.34	41.79	37.82	53.28	40.66	87.88
34	jonatasgrosman/wav2vec2-large-xlsr-53-arabic	0.3B	60.98	86.82	23.00	42.75	64.27	56.29	92.72
35	speechbrain/asr-wav2vec2-commonvoice-14-ar	0.1B	65.74	88.54	29.17	49.10	69.57	64.37	93.68

Bold = our model and our best-in-column results. lemura-arabic-asr-lite is the single best system on MASC-clean (7.27) and MASC-noisy (23.65) of every model evaluated, and the highest-ranked system below 1B parameters by a wide margin. Casablanca (Moroccan Darija) is the hardest set for every system.

Competitor rows are the published Open Universal Arabic ASR Leaderboard standings, reproduced for context; they are not our measurements.

Efficiency

The systems around us on the leaderboard are large generative audio-LLMs (2–30B parameters) that need GPUs. lemura-arabic-asr-lite reaches the same accuracy tier with a ~115M-parameter CTC model:

	lemura-arabic-asr-lite	Typical top-5 systems
Parameters	~115M	2B – 30B
Hardware	CPU or GPU	GPU
Latency	Real-time	Seconds / clip
Footprint	~0.4 GB	4 – 60 GB
Avg WER	25.08	24.78 – 30.71

That makes it practical for on-device, low-cost and high-throughput Arabic transcription where the larger models are impractical.

Inference

import nemo.collections.asr as nemo_asr

model = nemo_asr.models.ASRModel.restore_from("asr_final.nemo")
print(model.transcribe(["audio.wav"]))   # 16 kHz mono

Try it live, no install: lemura-arabic-asr demo

Languages, dialects and tasks

Languages	Arabic
Dialects	MSA, Gulf/Khaleeji, Egyptian, Levantine, Maghrebi
Tasks	Transcription (audio → text)
Audio	16 kHz mono, auto-resampled

Intended use and limitations

Intended: transcribing Arabic speech across dialects — call centres, voice notes, media captioning, voice interfaces, and any deployment where CPU-only or real-time operation matters.

Limitations:

• Maghrebi/Darija (Casablanca, 58.24 WER) remains hard, as it does for every system on the leaderboard.
• Very noisy or far-field audio degrades accuracy.
• CTC output has no built-in punctuation or diacritic restoration.
• May reflect biases present in the training corpora.

Related model

For a larger, generative alternative with broader dialect fine-tuning, see lemura-arabic-asr-qwen3 (1.7B, audio-LLM). Note its published WER is an in-domain figure and is not directly comparable to the zero-shot leaderboard numbers above.

License

CC-BY-4.0.

Citation

@misc{lemura_arabic_asr_lite_2026,
  title  = {lemura-arabic-asr-lite: Compact Multi-Dialect Arabic Speech Recognition},
  author = {Lemura Labs},
  year   = {2026},
  url    = {https://huggingface.co/lemuralabs/lemura-arabic-asr-lite}
}

About Lemura Labs

Arabic-first, efficiency-first speech intelligence.

Lemura Labs builds compact, deployable speech and language models — accuracy at a size and cost that works outside the datacentre.