{"dataset":{"name":"Speech recognition accuracy on long dictation: one 4-minute recording against the same speech in short pieces","title":"Dictation Accuracy on Long Recordings, Tested","question":"Do speech recognition engines transcribe several minutes of continuous speech as accurately as the same speech handed over in short pieces?","url":"https://yapapp.cc/data/asr-long-dictation","measure":{"label":"Word error rate","unit":"percent","short":"%","decimals":2,"lowerIsBetter":true},"dimensions":[{"key":"engine","label":"Engine","values":["Parakeet TDT v3","Whisper large-v3-turbo","Apple SpeechTranscriber"]},{"key":"maker","label":"Maker"},{"key":"condition","label":"How the speech was handed over","values":["One recording of about 4 minutes","The same audio in pieces of 10 to 20 seconds"]},{"key":"examples","label":"Readings"},{"key":"speakers","label":"Speakers"},{"key":"language","label":"Language"}],"include":"In every chapter of the test-clean split of LibriSpeech-Long (Google DeepMind, 2024, from LibriSpeech's audiobook recordings), its first example of at least 3 minutes: 84 readings of 3 to 4 minutes by all 40 of the split's speakers, 5.45 hours of clean read English. Each reading is transcribed twice by each engine: as one recording, and cut at pauses into pieces of 10 to 20 seconds whose text is joined.","exclude":"Readings under 3 minutes (a chapter's tail), the split's noisier recordings (test-other), other languages, people dictating their own words, streaming as an app transcribes while you speak, and cloud engines, for now.","method":"Each engine transcribed every reading on a Mac with English given. The pieces end at the quietest tenth of a second between 10 and 20 seconds after each piece starts, so they split the speech at pauses, as someone dictating in short bursts would. Word error rate is computed against LibriSpeech's own transcript with the same normalization as our main benchmark, and the same alignment counts the words an engine left out and the longest run of them. Speakers, not readings, are the unit of the margins of error: intervals come from resampling speakers with all their readings, the two conditions are compared on the same readings, and a difference counts as measured only when its interval, allowing for the test's comparisons, excludes zero. Every file's checksum, cut point, reference and transcript are in the full record.","limitations":"Audiobook narrators reading clean text in a quiet room are easier to transcribe than a person composing an email aloud, and their pauses are regular. The test hands each engine a finished recording; an app that transcribes while you speak, or cuts the audio its own way, may do better or worse than either condition. LibriSpeech writes numbers as words, so an engine that writes digits is charged a little in both conditions alike: compare each engine with itself across the two conditions for what length does, and read the engine against engine comparisons with that in mind.","sources":0,"updated":"2026-10-25","license":"https://creativecommons.org/licenses/by/4.0/","credit":"Yap (2026). Speech recognition accuracy on long dictation: one 4-minute recording against the same speech in short pieces [Data set]. https://yapapp.cc/data/asr-long-dictation"},"rows":[{"row":"Apple SpeechTranscriber, in one recording","value":2.34,"low":2.01,"high":2.67,"as_written":"2.01 to 2.67","unit":"percent","engine":"Apple SpeechTranscriber","maker":"Apple","condition":"One recording of about 4 minutes","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 2.34 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over as one recording of 3 to 4 minutes.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-apple-whole-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"},{"row":"Parakeet TDT v3, in one recording","value":2.7,"low":2.36,"high":3.06,"as_written":"2.36 to 3.06","unit":"percent","engine":"Parakeet TDT v3","maker":"NVIDIA","condition":"One recording of about 4 minutes","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 2.70 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over as one recording of 3 to 4 minutes.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-parakeet-whole-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"},{"row":"Whisper large-v3-turbo, in one recording","value":2.91,"low":2.54,"high":3.27,"as_written":"2.54 to 3.27","unit":"percent","engine":"Whisper large-v3-turbo","maker":"OpenAI","condition":"One recording of about 4 minutes","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 2.91 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over as one recording of 3 to 4 minutes.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-whisper-whole-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"},{"row":"Apple SpeechTranscriber, in pieces","value":2.33,"low":2.01,"high":2.65,"as_written":"2.01 to 2.65","unit":"percent","engine":"Apple SpeechTranscriber","maker":"Apple","condition":"The same audio in pieces of 10 to 20 seconds","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Apple SpeechTranscriber had a word error rate of 2.33 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over cut at pauses into pieces of 10 to 20 seconds.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-apple-pieces-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"},{"row":"Whisper large-v3-turbo, in pieces","value":2.35,"low":2.08,"high":2.63,"as_written":"2.08 to 2.63","unit":"percent","engine":"Whisper large-v3-turbo","maker":"OpenAI","condition":"The same audio in pieces of 10 to 20 seconds","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 2.35 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over cut at pauses into pieces of 10 to 20 seconds.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-whisper-pieces-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"},{"row":"Parakeet TDT v3, in pieces","value":2.45,"low":2.1,"high":2.83,"as_written":"2.1 to 2.83","unit":"percent","engine":"Parakeet TDT v3","maker":"NVIDIA","condition":"The same audio in pieces of 10 to 20 seconds","examples":84,"speakers":40,"language":"English","conversion":null,"claim":"In our October 2026 test on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 2.45 percent on 84 English readings of 3 to 4 minutes by 40 speakers, handed over cut at pauses into pieces of 10 to 20 seconds.","quote":null,"where_in_source":null,"fact_url":"https://yapapp.cc/facts/m-longform-2026-10-parakeet-pieces-wer","source_title":null,"source_publisher":null,"source_year":null,"source_url":null,"source_doi":null,"checked":"2026-10-25"}]}