Cohere-Speech-Tashkeel-2B

Home/Models/NAMAA-Space/Cohere-Speech-Tashkeel-2B

NAMAA-Space/Cohere-Speech-Tashkeel-2B

kind: original
quant: bf16 (16-bit)
family: command
modality: automatic-speech-recognition
root model: CohereLabs/cohere-transcribe-arabic-07-2026
provider: NAMAA-Space
license: apache-2.0
language: ar
context: —
library: transformers
vRAM (GB): —
size on disk: —
downloads: 2,425
likes: 22
created: —
last modified: 2026-07-13T11:11:49+00:00 (2026)
huggingface: https://huggingface.co/NAMAA-Space/Cohere-Speech-Tashkeel-2B
tags: transformerssafetensorscohere_asrautomatic-speech-recognitionspeech-to-textarabicdiacritizationtashkeelharakatiraabcoherearbase_model:CohereLabs/cohere-transcribe-arabic-07-2026base_model:finetune:CohereLabs/cohere-transcribe-arabic-07-2026license:apache-2.0model-indexendpoints_compatibleregion:us

benchmark	category	score	context	conf	evidence
WER	speech-asr	24.35	WER — fully diacritized	medium	`\| WER — fully diacritized \| 24.35% \| Strict: any single wrong ḥaraka marks the whole word wrong. \|`
WER	speech-asr	10.13		low	`\| 95.30% \| 6.62% \| 5.67% \| 10.13% \|`
WER	speech-asr	10.13	WER — undiacritized	medium	`\| WER — undiacritized \| 10.13% \| Letter-level speech-recognition accuracy. \|`

none scraped.