Malaysian speech to text benchmark

Aisyah 1.0 Pro vs YTL ILMU ASR v4.2: Malaysian speech to text compared

We ran Aisyah 1.0 Pro and YTL ILMU ASR v4.2 on the same 1,899 Malaysian speech clips, from telephony and street interviews to parliament and singing, and compared their word error rates.

Aisyah 1.0 Pro
Fewer mistakes overall, and on all 12 kinds of audio. Biggest leads: singing and short replies.
YTL ILMU ASR v4.2
Not ahead of Aisyah 1.0 Pro on any kind of audio, noise level or clip set. Closest on podcasts: 6.01% against 4.57%.

Revolab built this benchmark and Aisyah. Results as of 12 August 2026. How we tested

Word error rate on all 1,899 clipsMistakes per 100 words spoken. Shorter bar, fewer mistakes.
  • Aisyah 1.0 Pro5.79%
  • YTL ILMU ASR v4.29.48%

Where Aisyah and ILMU ASR v4.2 each make fewer mistakes

Aisyah 1.0 ProYTL ILMU ASR v4.2Word error rate. Further left, fewer mistakes.

Aisyah 1.0 Pro makes fewer mistakes on 12 of 12

  • Singing152 clips
    Aisyah 1.0 Pro 11.01%YTL ILMU ASR v4.2 28.80%
  • Short repliesShort inputs, 155 clips
    Aisyah 1.0 Pro 3.28%YTL ILMU ASR v4.2 20.17%
  • Phone callsTelephony, 229 clips
    Aisyah 1.0 Pro 9.55%YTL ILMU ASR v4.2 20.40%
  • Drama156 clips
    Aisyah 1.0 Pro 5.91%YTL ILMU ASR v4.2 9.25%
  • Animation150 clips
    Aisyah 1.0 Pro 5.66%YTL ILMU ASR v4.2 8.84%
  • Street interviews143 clips
    Aisyah 1.0 Pro 14.24%YTL ILMU ASR v4.2 17.02%
  • Scripted readingRead speech, 154 clips
    Aisyah 1.0 Pro 1.30%YTL ILMU ASR v4.2 3.80%
  • Common Voicevolunteer read-aloud, 141 clips
    Aisyah 1.0 Pro 3.74%YTL ILMU ASR v4.2 6.15%
  • FLEURSread-aloud research set, 153 clips
    Aisyah 1.0 Pro 3.42%YTL ILMU ASR v4.2 5.69%
  • Parliament155 clips
    Aisyah 1.0 Pro 3.79%YTL ILMU ASR v4.2 5.36%
  • News156 clips
    Aisyah 1.0 Pro 3.48%YTL ILMU ASR v4.2 5.01%
  • PodcastsPodcast, 155 clips
    Aisyah 1.0 Pro 4.57%YTL ILMU ASR v4.2 6.01%

ILMU ASR v4.2 makes fewer mistakes on 0 of 12

No kind of audio in this benchmark. It comes closest on podcasts: 6.01% against 4.57%.

By background noise and clip set

The same 1,899 clips, split another way. The lower figure is in bold.

  • Clean audio1,364 clips
    Aisyah 1.0 Pro 4.70%YTL ILMU ASR v4.2 7.79%
  • Some background noise277 clips
    Aisyah 1.0 Pro 6.12%YTL ILMU ASR v4.2 8.93%
  • Noisy audio249 clips
    Aisyah 1.0 Pro 12.16%YTL ILMU ASR v4.2 20.91%
  • Published clips820 clips anyone can check
    Aisyah 1.0 Pro 4.89%YTL ILMU ASR v4.2 7.78%
  • Held-back clips1,079 clips no model has seen
    Aisyah 1.0 Pro 6.58%YTL ILMU ASR v4.2 10.97%

Hear the difference

Real clips from the published half of the benchmark. Play the audio, then read what each model returned, unedited.

A ringgit amount in words

Phone calls, 2.9 s
What was said
I dah bayar lah, RM950 ni.
Aisyah 1.0 Pro
I dah bayar lah, sembilan ratus lima puluh ringgit ni.
YTL ILMU ASR v4.2
i dah bayar lah sembilan ratus lima puluh ringgit ni
What to listen for

Aisyah 1.0 Pro and YTL ILMU ASR v4.2 returned the same words here, both writing "RM950" as "sembilan ratus lima puluh ringgit". Only capitals and punctuation differ between the two outputs.

Short eligibility question

Phone calls, 1.9 s
What was said
Credit card am I eligible?
Aisyah 1.0 Pro
Credit card am I eligible?
YTL ILMU ASR v4.2
Credit card emi eligible
What to listen for

Aisyah 1.0 Pro returned "Credit card am I eligible?", the same as the reference. YTL ILMU ASR v4.2 returned "Credit card emi eligible", with "emi" where the reference has "am I".

Street interview in Manglish

Street interviews, 4.9 s
What was said
Apa bagus dia ada yang tak bagus dia. Depends lah. Depends pada individu. Haa, aa ah, individu lah.
Aisyah 1.0 Pro
Apa bagus dia, ada yang tak bagus dia. depends lah. depends pada individu lah.
YTL ILMU ASR v4.2
Apa bagus dia ada yang tak bagus dia. Depends lah. Haah individu lah
What to listen for

For the stretch the reference gives as "Depends pada individu. Haa, aa ah, individu lah.", Aisyah 1.0 Pro returned "depends pada individu lah." and YTL ILMU ASR v4.2 returned "Haah individu lah".

Aisyah vs YTL ILMU, answered.

Is Aisyah more accurate than YTL ILMU?

On Revolab's Malaysian speech benchmark of 1,899 clips, Aisyah 1.0 Pro has a word error rate of 5.79% and YTL ILMU ASR v4.2 has 9.48%. Aisyah 1.0 Pro has the lower WER in all 12 categories and on clean, moderate and noisy audio. Revolab built both the benchmark and Aisyah.

How accurate is YTL ILMU ASR on telephony audio?

On the 229 Telephony clips in Revolab's Malaysian speech benchmark, YTL ILMU ASR v4.2 has a word error rate of 20.40% and Aisyah 1.0 Pro has 9.55%, a gap of 10.85 percentage points. Lower is better.

Is YTL ILMU ASR better than Aisyah in any category?

No. On Revolab's Malaysian speech benchmark, YTL ILMU ASR v4.2 does not have a lower WER than Aisyah 1.0 Pro in any of the 12 categories, at any noise level, or on the 820 published or 1,079 held-back clips. The narrowest category gap is Podcast, at 1.44 percentage points.

Which is more accurate on noisy audio, Aisyah or YTL ILMU?

On the 249 clips tagged noisy in Revolab's Malaysian speech benchmark, Aisyah 1.0 Pro has a word error rate of 12.16% and YTL ILMU ASR v4.2 has 20.91%, a gap of 8.75 percentage points. Aisyah 1.0 Pro also has the lower WER on clean and moderate clips.

Check the numbers yourself.

Revolab built this benchmark and trained Aisyah, so read the results with that in mind. Every published clip and every model's output is open to check.

All the numbersEvery figure for all 2 models on this page, as tables

Overall

Word error rate for each model. Lower is better; the lowest in each column is highlighted.
ModelAll clips1,899 clipsPublished820 clipsHeld back1,079 clips
Aisyah 1.0 Pro5.79% (fewer mistakes)4.89% (fewer mistakes)6.58% (fewer mistakes)
YTL ILMU ASR v4.29.48%7.78%10.97%

By kind of audio

Word error rate by kind of audio. Lower is better; the lowest in each row is highlighted.
Kind of audioAisyah 1.0 ProILMU ASR v4.2
Phone callsTelephony, 229 clips9.55% (fewer mistakes)20.40%
Short repliesShort inputs, 155 clips3.28% (fewer mistakes)20.17%
Scripted readingRead speech, 154 clips1.30% (fewer mistakes)3.80%
PodcastsPodcast, 155 clips4.57% (fewer mistakes)6.01%
Drama156 clips5.91% (fewer mistakes)9.25%
Animation150 clips5.66% (fewer mistakes)8.84%
News156 clips3.48% (fewer mistakes)5.01%
Parliament155 clips3.79% (fewer mistakes)5.36%
Street interviews143 clips14.24% (fewer mistakes)17.02%
Singing152 clips11.01% (fewer mistakes)28.80%
Common Voicevolunteer read-aloud, 141 clips3.74% (fewer mistakes)6.15%
FLEURSread-aloud research set, 153 clips3.42% (fewer mistakes)5.69%

By background noise

Word error rate by background noise. Lower is better; the lowest in each row is highlighted.
BackgroundAisyah 1.0 ProILMU ASR v4.2
Clean audio1,364 clips4.70% (fewer mistakes)7.79%
Some background noise277 clips6.12% (fewer mistakes)8.93%
Noisy audio249 clips12.16% (fewer mistakes)20.91%

Types of mistake

Types of mistake per 100 words spoken. Lower is better.
ModelWrong wordsper 100 wordsInvented wordsper 100 wordsDropped wordsper 100 words
Aisyah 1.0 Pro3.48% (fewer mistakes)0.62% (fewer mistakes)1.69% (fewer mistakes)
YTL ILMU ASR v4.26.15%0.87%2.46%
How the test worksClips, scoring, noise tags and what is not compared
  • 1,899 Malaysian utterances in 12 kinds of audio, from phone calls and street interviews to parliament and singing. 820 are published and 1,079 are held back.
  • Word error rate counts every wrong, invented and dropped word and divides by the number of words actually said, across all clips. Lower is better.
  • Transcripts are normalised before scoring, so casing, punctuation and writing a number as digits or as words do not count as errors.
  • Clips are tagged for background noise: 1,364 clean, 277 moderate and 249 noisy.
  • Speed is not compared. For models served over an API, the time measured is mostly network latency rather than the model itself.
  • Results as of 12 August 2026, under the model names shown. Vendors update their models, so results can change.

Product and company names are trademarks of their respective owners and are used only to identify the models tested. Revolab is not affiliated with or endorsed by any of them.

Compare Aisyah with other models

See every model in one table

Test it on your own audio.

Every new account gets 50 free minutes of speech to text and 50 of text to speech. No card required.