Glossary
The Lombard effect is the involuntary speech change humans (and some animals) produce in noisy environments: louder volume, higher pitch, slower speech, more deliberate articulation, longer vowels. Discovered by French otolaryngologist Étienne Lombard in 1909. It's why people shout in nightclubs even when their conversation partner is right next to them. Lombard speech sounds different enough that ASR models trained only on quiet-environment speech degrade on it.
Acoustic features of Lombard speech
Compared to neutral conversational speech, Lombard speech has: +5-10 dB volume, +10-15% fundamental frequency, +30-100% longer vowel durations, more energy in mid/high frequencies (2-4 kHz where speech intelligibility is concentrated), and reduced spectral tilt. Speakers also produce fewer phoneme-level reductions — "going to" stays as "going to" instead of becoming "gonna".
Why it matters for ASR
If your ASR model was trained primarily on clean studio recordings (audiobooks, interviews in quiet rooms), it has never seen Lombard speech. Test it on a noisy bar recording — even after you remove the background noise — and accuracy drops 5-10 WER points. The speech is structurally different, not just noisier. Modern multi-condition training (including Lombard examples in the training set) fixes this. Whisper was trained on 680,000 hours of varied audio including plenty of Lombard speech — it handles it well.
Practical implications
If you're recording in a noisy environment, the Lombard effect is your friend — speakers naturally articulate more clearly, partially compensating for the noise. Use a directional or close-mic'd microphone to reduce the noise reaching the recording. Whisper handles the rest. If accuracy is still poor, the problem is likely the noise itself (overwhelming the speech), not the Lombard adaptation.
In practice
Record an interview in a busy coffee shop with your phone on the table. The speakers will naturally apply Lombard adaptations — louder, slower, clearer than conversational speech in a quiet room. Whisper-1 typically achieves 90-93% accuracy on such recordings, surprisingly good given the conditions. Compare to recording in a podcast studio at the same pace: accuracy 95-97%.
Related terms
Further reading
Transcribe noisy audio
Get started — freeLast updated: September 20, 2026
Frequently Asked Questions
What does Lombard effect mean in transcription?
The involuntary tendency of speakers to raise their voice and articulate more clearly in noisy environments. Recordings made under high noise sound louder than normal conversation, which complicates ASR if the model wasn't trained on Lombard speech.
Why does Lombard effect matter when choosing a transcription workflow?
Lombard effect affects how you understand transcript quality, timing, compatibility, or the technology behind speech-to-text results.
Where can I apply Lombard effect on Transcript.you?
Use the related workflow at /ai-transcription when you want to see how this glossary concept connects to an actual transcription task.
Browse the full transcription glossary or read the complete guide to AI transcription.