In our October 2026 test on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 6.84 percent on 42 Italian readings of a short passage by 21 healthy speakers aged 60 to 77.

Measured by us on October 23, 2026, from 42 recordings

Our own measurement

This number comes from our benchmark, not from another publication. The full record (the setup, every recording's checksum, every transcript and timing) is published, so you can recompute it.

Dictation Accuracy With Parkinson's Speech, Tested Download the full record (JSON)

What it measures: OpenAI Whisper large-v3-turbo, 5-bit quantized (ggml-large-v3-turbo-q5_0.bin), run through whisper.cpp 1.9.4 with Metal on the GPU, with the language given. Whisper ran with greedy decoding and timestamps on, the way whisper.cpp moves through recordings longer than its 30-second window (most of these are 50 to 250 seconds). A first pass with timestamps off lost most of several recordings and is not used. whisper.cpp's beam search was not tested. The speakers read a printed text aloud in a quiet room with a microphone 15 to 25 cm away (Italian Parkinson's Voice and Speech, University of Bari and ASL Bari, recorded 2016 and 2017). The reference is the text they were given, so a word a speaker skipped or changed counts as an error for every engine alike.
How it is measured: Word error rate = 370 word errors / 5,410 reference words = 6.84 percent, over 42 recordings (46.7 minutes of audio). The 95% interval over speakers is 4.97 to 9.22 percent.
What it does not show: Against healthy speakers aged 19 to 29 reading the same text, this is +3.80 points; the 95% interval for that difference, allowing for the test's 9 comparisons, is +1.21 to +7.52 points, which excludes zero, so this is a measured difference. 21 speakers from one region of Italy, in one language, reading a fixed text: a person dictating their own words, in another language or with more advanced Parkinson's may fare differently. Half of these speakers had an error rate of 5.4 percent or less, the highest was 23.3 percent.

In the data

A row of Dictation Accuracy With Parkinson's Speech, Tested, as Whisper large-v3-turbo, Healthy speakers aged 60 to 77, read passage: 4.97 to 9.22 percent.

Engine: Whisper large-v3-turbo
Maker: OpenAI
Speakers: Healthy speakers aged 60 to 77
What they read: Read passage
Speakers in group: 21
Recordings: 42
Language: Italian

Used to calculate

The speech recognition accuracy on the speech of people with Parkinson's disease dataset holds 15 data points, each a published figure traced to its source.

Our own measurement

In the data

Used to calculate

Cited on