oddadmix

qwen3-asr-1.7b-arabic-dialectal-v2

README

License: apache-2.0

Results (balanced test split ~2,600 clips, `clean_text` WER/CER)

Table with columns: WER, CER
	WER	CER
This model (v3)	0.478	0.182

Per-dialect accuracy (v3 balanced test, `clean_text`)

WER / CER on each dialect's 200 held-out clips, easiest to hardest. Gulf/Levantine transcribe best; Maghrebi (Tunisian/Algerian/Moroccan) is hardest — consistent across every model.

Table with columns: Dialect, WER, CER, n
Dialect	WER	CER	n
Saudi/Gulf (`sa`)	0.253	0.072	200
Egyptian (`eg`)	0.322	0.115	200
Iraqi (`iq`)	0.340	0.101	200
Bahraini (`bh`)	0.424	0.165	200
Syrian (`sy`)	0.432	0.179	200
Yemeni (`ye`)	0.456	0.180	200
Palestinian (`ps`)	0.461	0.177	200
Sudanese (`sd`)	0.511	0.209	200
Lebanese (`lb`)	0.531	0.215	200
Libyan (`ly`)	0.553	0.220	200
Moroccan (`ma`)	0.556	0.187	200
Algerian (`dz`)	0.576	0.203	200
Tunisian (`tn`)	0.824	0.363	200
Overall	0.478	0.182	2600

v2 comparison — all models on the balanced dataset

Same dialect-balanced test split (~2,600 clips, oddadmix/lahgtna-v3-small), same clean_text scoring, seed 42.

Table with columns: Model, Params, WER, CER
Model	Params	WER	CER
whisper-large-v3-v2	1.54B	0.320	0.120
whisper-large-v3-turbo-v2	809M	0.332	0.125
whisper-medium-v2	769M	0.342	0.128
whisper-small-v2	244M	0.403	0.153
nemotron-3.5-asr-v2 (streaming)

whisper-large-v3-turbo-v2 is the best; Whisper fine-tunes furthest. nemotron is the streaming option. (v1 models were trained on the earlier augmentation-expanded set.)

Examples — fine-tuned output (one clip per dialect)

13 held-out test clips, one per dialect, transcribed by fine-tuned (v3) (this model). Text is clean_text-normalized (undiacritized), matching the WER scoring. Press play to listen.

Table with columns: Dialect, Audio, Reference, Fine-tuned
Dialect	Reference	Fine-tuned
Egyptian (`eg`)	لا، في كافيه هنا كده جنب الدار، مع أحمد خطيبها، وزمانها جاي.	لا في كافي هنا كده جنب الدار، مع أحمد خطبها و زمانها جاي.
Lebanese (`lb`)	ما كون مأمنة حالي. ما أقدر اتحكم. امم. في عندي خوف، إيه.	ما كون مأمني حالي، ما قدرت اتحكم، في عندي خوف إيه.
Syrian (`sy`)	يعني إنت كان في عندك نظام فاقد الشرعية الدولية، دول ما كانت تعترف فيه. إقامة و دول

Base-column transcripts (Qwen/Qwen3-ASR-1.7B-hf) omitted — base weights weren't available locally.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

oddadmix

Model Tree

Base

Qwen/Qwen3-ASR-1.7B-hf

Fine-tuned

this model

Input Modalities

Audio

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Results (balanced test split ~2,600 clips, `clean_text` WER/CER)

Table with columns: WER, CER
	WER	CER
This model (v3)	0.478	0.182

Per-dialect accuracy (v3 balanced test, `clean_text`)

WER / CER on each dialect's 200 held-out clips, easiest to hardest. Gulf/Levantine transcribe best; Maghrebi (Tunisian/Algerian/Moroccan) is hardest — consistent across every model.

Table with columns: Dialect, WER, CER, n
Dialect	WER	CER	n
Saudi/Gulf (`sa`)	0.253	0.072	200
Egyptian (`eg`)	0.322	0.115	200
Iraqi (`iq`)	0.340	0.101	200
Bahraini (`bh`)	0.424	0.165	200
Syrian (`sy`)	0.432	0.179	200
Yemeni (`ye`)	0.456	0.180	200
Palestinian (`ps`)	0.461	0.177	200
Sudanese (`sd`)	0.511	0.209	200
Lebanese (`lb`)	0.531	0.215	200
Libyan (`ly`)	0.553	0.220	200
Moroccan (`ma`)	0.556	0.187	200
Algerian (`dz`)	0.576	0.203	200
Tunisian (`tn`)	0.824	0.363	200
Overall	0.478	0.182	2600

v2 comparison — all models on the balanced dataset

Same dialect-balanced test split (~2,600 clips, oddadmix/lahgtna-v3-small), same clean_text scoring, seed 42.

Table with columns: Model, Params, WER, CER
Model	Params	WER	CER
whisper-large-v3-v2	1.54B	0.320	0.120
whisper-large-v3-turbo-v2	809M	0.332	0.125
whisper-medium-v2	769M	0.342	0.128
whisper-small-v2	244M	0.403	0.153
nemotron-3.5-asr-v2 (streaming)

whisper-large-v3-turbo-v2 is the best; Whisper fine-tunes furthest. nemotron is the streaming option. (v1 models were trained on the earlier augmentation-expanded set.)

Examples — fine-tuned output (one clip per dialect)

13 held-out test clips, one per dialect, transcribed by fine-tuned (v3) (this model). Text is clean_text-normalized (undiacritized), matching the WER scoring. Press play to listen.

Table with columns: Dialect, Audio, Reference, Fine-tuned
Dialect	Reference	Fine-tuned
Egyptian (`eg`)	لا، في كافيه هنا كده جنب الدار، مع أحمد خطيبها، وزمانها جاي.	لا في كافي هنا كده جنب الدار، مع أحمد خطبها و زمانها جاي.
Lebanese (`lb`)	ما كون مأمنة حالي. ما أقدر اتحكم. امم. في عندي خوف، إيه.	ما كون مأمني حالي، ما قدرت اتحكم، في عندي خوف إيه.
Syrian (`sy`)	يعني إنت كان في عندك نظام فاقد الشرعية الدولية، دول ما كانت تعترف فيه. إقامة و دول

Base-column transcripts (Qwen/Qwen3-ASR-1.7B-hf) omitted — base weights weren't available locally.

qwen3-asr-1.7b-arabic-dialectal-v2

README

Results (balanced test split ~2,600 clips, clean_text WER/CER)

Per-dialect accuracy (v3 balanced test, clean_text)

v2 comparison — all models on the balanced dataset

Examples — fine-tuned output (one clip per dialect)

Explore FriendliAI today

README

Results (balanced test split ~2,600 clips, clean_text WER/CER)

Per-dialect accuracy (v3 balanced test, clean_text)

v2 comparison — all models on the balanced dataset

Examples — fine-tuned output (one clip per dialect)

Results (balanced test split ~2,600 clips, `clean_text` WER/CER)

Per-dialect accuracy (v3 balanced test, `clean_text`)

Results (balanced test split ~2,600 clips, `clean_text` WER/CER)

Per-dialect accuracy (v3 balanced test, `clean_text`)