At 128 kbps an MP3 transcribed word-for-word the same as the 320 kbps master; at 64 kbps 0.65% of words changed, and at 32 kbps 7.26% — and the words that changed were the names. Meanwhile a recording cut down to genuine telephone quality came back with an identical transcript. Those two tests, plus the bitrates of 10,306 real MP3 uploads, are what the second episode of The Transcript Lab is about.
Free
The Transcript Lab is a short podcast from TranscribeNext about what transcription actually does, measured rather than assumed. Every number in the episode comes from data we collected ourselves and publish in full.
Listen and read along
The player below runs the episode next to its transcript, produced by TranscribeNext from this exact MP3. The highlighted word is the one the engine timed at that moment. Click any line to jump to it.
The Transcript Lab · Episode 2
How bad can your audio be? · 5:08
0:00 / 5:08
The transcript is what TranscribeNext returned for this exact MP3, with speaker identification on (two speakers) and the speaker labels renamed to the hosts. Nothing else is edited: every line kept the speaker the engine gave it, and numbers are written as digits because that is how the engine wrote them. Line and word timings are measured, so the highlight follows the words as the engine timed them.
What the episode covers
The bitrate test: one five-minute recording encoded from the same master at 320, 128, 64 and 32 kbps, every transcript compared word by word
128 kbps: identical, 620 words out of 620. 64 kbps: 0.65% different. 32 kbps: 7.26%, with "Milo" heard as "Michael" and "artillery" as "art told me"
Real files: across 10,306 MP3 uploads the median is 128 kbps, but 54.1% are below it and 22.9% — about one in five — are below 64 kbps
The phone test: a clean recording re-encoded to 8 kHz mono, 300–3400 Hz, mu-law. The words came back identical; only the segmentation changed, from 7 segments to 3
What to do about it: upload the original file, treat 64 kbps as fine for speech, and check names and numbers on low-bitrate recordings
The bitrate test and the real-upload data, with the full method, are in the guide to the best MP3 to text converter. The phone-quality test and where voicemail transcripts really go wrong are in the voicemail to text guide.
