Model Reviews

Qwen3-ASR-1.7B: Speech-to-Text in 6 Audio Tests

Qwen3-ASR-1.7B: Speech-to-Text in 6 Audio Tests

Qwen3-ASR-1.7B is a lightweight speech-to-text model on Wiro. This post runs 6 repeatable audio tests and shows the transcripts it returned.

Model link: https://wiro.ai/models/qwen/qwen3-asr-1-7b

What was tested

  • English dictation with numbers and a tracking code
  • English dictation with URL and token-like strings
  • Turkish e-commerce style sentence
  • Spanish customer support style sentence
  • The same English dictation with added white noise
  • The token-heavy English clip sped up 1.35x

Test setup

All input audio clips were generated with https://wiro.ai/models/resemble-ai/chatterbox-multilingual so the scripts stay consistent across reruns. Then the audio was uploaded to WordPress and used as the ASR input URL. Qwen3-ASR-1.7B was run with language selection plus batchSize=32 and newTokens=256.

Results

Test Language Condition Notes
1 English Clean Strong punctuation and numbers
2 English Clean Struggled with domain and token spelling
3 Turkish Clean Major errors in amounts and codes
4 Spanish Clean Digits and time were unstable
5 English White noise mix Held up well on this clip
6 English 1.35x speed Similar errors to test 2

Test 1: English clean dictation

Prompt script: For the shipping audit, order 48219 shipped on February 14 at 9:05 AM. Total weight 3.7 kilograms. Tracking code Z X dash 9 1 dash Delta.

Qwen3-ASR output:

For the shipping audit, order 48219 shipped on February 14 at 9:05 a.m. Total weight 3.7 kilograms. Tracking code ZX-91-DELTA.

Test 2: English with URL and token-like strings

Prompt script: Email support plus wiro at acme dot dev. URL https colon slash slash api dot example dot com slash v1 slash run question mark mode equals fast ampersand retry equals 2. Error code E underscore C O N N underscore R E S E T. Commit seven f three a nine c one.

Qwen3-ASR output:

Email support plus yiro at acme.dev. URL s colon slash slash api.example.com slash v1 slash run question mark mode equals fast ampersand retry equals two. Error code e underscore c o n n underscore r e s e t. Commit seven f three a nine c one.

Test 3: Turkish

Prompt script: Sepet tutari 1.249,90 TL. Kargo kodu T R dash 508 dash A B. Teslimat 3 gun icinde. Iade suresi 14 gundur.

Qwen3-ASR output:

Sepet tutarı 180000 komisiki 16 TL kargo kodu TR016

Test 4: Spanish

Prompt script: El pedido numero 1740 llego el martes a las 18:30. El codigo postal es 28013. Gracias por llamar.

Qwen3-ASR output:

El pedido número B 740 llegó el martes a las de 8:40. El código postal es 2800 S. Gracias por llamar. Hay fiends en tu.

Test 5: English with added white noise

Prompt script: Same as test 1, mixed with white noise.

Qwen3-ASR output:

For the shipping audit, order 48219 shipped on February 14 at 9:05 a.m. Total weight 3.7 kilograms. Tracking code ZX-91-DELTA.

Test 6: English token clip at 1.35x speed

Prompt script: Same as test 2, audio sped up 1.35x.

Qwen3-ASR output:

Email support plus wireo at acme.dev. URL s colon slash slash api.example.com slash v1 slash run question mark mode equals fast ampersand retry equals two error code e underscore c o n n underscore r e s e t commit seven f three a nine c one

Quick takeaways

  • Clean English dictation looked solid, including numbers and punctuation.
  • Token-heavy text (URLs, underscores, commit hashes) needs cleanup rules on the client side.
  • This Turkish and Spanish sample did not hold up. More language-specific testing is needed before relying on it.

Try it

Qwen3-ASR-1.7B on Wiro


Leave a Comment

Your email address will not be published. Required fields are marked *