Glossary
LibriSpeech
LibriSpeech is a free corpus of about 1,000 hours of read English speech, taken from LibriVox audiobooks, and it is one of the most widely used benchmarks for English speech recognition. Its scores describe clean, read speech, which differs from everyday dictation.
Updated October 17, 20263 cited facts, each traced to its source
LibriSpeech is a free corpus of speech and text, built for training and testing speech recognition systems. It holds about 1,000 hours of 16 kHz English speech, so researchers use it as a common yardstick: when two engines are both scored on LibriSpeech, their error rates can be compared on the same recordings.
It is built from read audiobooks
The corpus was assembled from recordings in the LibriVox project, a volunteer audiobook project. The data is derived from read audiobooks, and the speech is carefully segmented and aligned to the book texts 1. Each segment therefore has a reference transcript that a recognizer's output can be checked against, word by word.
The official test sets are split into a "clean" set, with easier recordings, and an "other" set, with harder recordings. Results are usually reported for both, because a system that scores well on clean speech can still struggle on the other set.
Its scores are a standard, not a promise
Scores on LibriSpeech are measured in word error rate. On the clean test set, Whisper Large V2 had a word error rate of 2.7 percent 2, which is 2.7 wrong words for every 100 in the reference. The number is a useful comparison between engines, but it describes read English, recorded with a good microphone, and it does not predict how a system will behave in your own room.
There is also a caution about how the data is used. A 2025 study found that a substantial amount of the LibriSpeech and Common Voice evaluation sets appear in public large language model pretraining corpora 3. If a model has seen the test recordings or their text during training, its score can overstate what it does on new speech. Treat a leaderboard number as one input, and check the same engine on your own audio.
Why it matters for dictation
Dictation is not read audiobook speech. You speak in bursts, mid-sentence, with names, product names and jargon, in rooms with fans, keyboards and traffic. LibriSpeech has none of those conditions in its clean set, so a low score there says little about how an engine handles:
- Proper names and technical terms that are rare in books.
- Spoken commands such as "new line" or "comma".
- Numbers, dates and symbols, which the test text does not require you to format.
- Background noise and a laptop microphone at arm's length.
For a dictation app, the more useful question is how the word error rate changes on recordings like yours. See how accurate speech recognition is in practice for the numbers that come from harder test sets and real recordings.
How LibriSpeech is used in practice
Many research papers on English recognition report results on the LibriSpeech test sets, and several open-source engines publish their scores on them too. Benchmark pages rank systems on the same recordings, with a normalized form of the text. Read how speech recognition benchmarks and leaderboards work to see what the rankings include and what they leave out.
Related terms
- Word error rate (WER): the measure used to score a transcript against its reference.
- Automatic speech recognition (ASR): the technology that LibriSpeech is used to test.
- Acoustic model: the part of a recognizer that is trained on corpora like this one.
- Mel spectrogram: the form of audio features that most recognizers compute before they read the speech.
Guides that cover it
Evidence
Every number on this page, with where it comes from. Open a fact to see the exact words in its source.
- 1.
- 2.
- 3.
Related terms
- Speech Recognition: What It Is and How Well It WorksThe topic this term belongs to
- Acoustic modelAn acoustic model is the part of a speech recognizer that scores how well a short window of audio matches the sounds of speech. Since about 2012 most acoustic models are deep neural networks, and their training data strongly affects accuracy.
- Automatic speech recognition (ASR)Automatic speech recognition (ASR) is software that converts spoken audio into written words. Modern systems are neural networks trained on large amounts of transcribed speech, and they run in the cloud or on the device.
- ConformerA Conformer is a neural encoder for speech recognition that combines self-attention, which sees the whole input, with convolution, which captures local sound patterns. FastConformer is NVIDIA's faster variant of the same design, used in streaming models.
- Connectionist temporal classification (CTC)Connectionist temporal classification (CTC) is a way to train a neural network to turn audio into text without knowing in advance which moment of sound matches which letter. The network labels every short slice of audio with a character or a blank, and the blanks and repeated labels are then removed to leave the transcript.
- Hidden Markov model (HMM)A hidden Markov model (HMM) is a statistical model in which a sequence of hidden states produces the observations you can measure, and each state changes to the next with set probabilities. In speech recognition, the hidden states are units of sound and the observations are short frames of audio.
- Inverse text normalization (ITN)Inverse text normalization (ITN) is the step that converts a speech recognizer's spoken-form output into written form, so that 'two hundred' becomes 200. It changes how the text reads, not which words the recognizer heard.
- Language model (in speech recognition)In speech recognition, a language model (LM) scores how probable a word sequence is in a language. The recognizer combines that score with the acoustic evidence to choose between words that sound alike, such as homophones.
- Mel spectrogramA Mel spectrogram is a grid of sound energy over time, with frequency bands spaced to match human hearing. Many speech recognizers compute its logarithm, the log-Mel spectrogram, and feed it to a neural network instead of the raw waveform.
- Real-time factor (RTF)The real-time factor (RTF) is the time a system takes to process audio divided by the length of that audio. An RTF of 1 or lower means the system keeps up with real-time speech. Some leaderboards report the inverse, RTFx, where higher is faster.
- RNN transducer (RNN-T)An RNN transducer (RNN-T) is a neural speech model that reads audio in order and emits words as it goes, so text can appear while the speaker is still talking. It has an encoder for the sound, a prediction network for the words already written, and a joint network that chooses the next output.
- Speaker diarizationSpeaker diarization is the task of working out who spoke when in a recording, by splitting the audio into turns and grouping turns that come from the same voice. It is separate from transcribing the words, and a transcript gains speaker labels only when diarization is run alongside recognition.
- Voice activity detection (VAD)Voice activity detection (VAD) is a component that decides whether a stretch of audio contains speech. Speech recognizers use it to split audio into turns, skip silence and decide when a speaker has finished.
Pages that use this term
- Speech Recognition: What It Is and How Well It WorksTopic
- What Is Whisper AI? OpenAI's Speech Model ExplainedGuide
- Word Error Rate (WER): How to Calculate and Read ItGuide
- How Accurate Is Speech Recognition? The Real NumbersGuide
- Speech Recognition Benchmarks and LeaderboardsGuide
- Speech Recognition in Python: Libraries ComparedGuide
- What are the best datasets for speech recognition?Question
- ConformerGlossary
- Mel spectrogramGlossary
- wav2vec 2.0Glossary