Vocational training — transcription phase of the same programme

Twenty-one thousand hours of Arabic dialects, transcribed on one office machine

A whole archive of recorded sessions in five Arabic dialect regions transcribed on a single graphics card in the office, with the model chosen by measurement, the audio never leaving the building, and no existing transcript overwritten.

One recording's path in five steps from left to right: the clearest audio is chosen, the call's own audio track before the video's sound; the speech is cut at pauses, never mid-word, with long sessions in fixed blocks; it is transcribed on one office graphics card with no cloud service; each segment is checked and empty, looping or garbled text is redone once; and the result is stored as a new version beside the old transcript, never over it. Below the transcription step: the model was chosen by testing on five dialect regions, with the hardest one setting the settings. Below the checking step, a dashed red branch: segments still flagged are counted per recording for a person to check. Below the last step: a person approves, and the team marks the version it trusts. A footer notes that it starts with the machine, restarts after a crash or a 30-minute stall, and resumes mid-recording.
The audio never leaves the office, the model was picked by measurement rather than reputation, and every transcript lands beside the old one for a person to choose.

Context

The same organisation as the earlier case studies in this programme. Its archive holds tens of thousands of recorded live sessions, in the dialects of five regions: Egypt and Sudan, the Gulf, Iraq and Yemen, the Levant, and the Maghreb. About 11,900 sessions had transcripts made one at a time with a general-purpose AI tool, without timings. Tens of thousands had none, so the only way to find a moment in a session was to watch it.

The problem

There was no answer key. Every existing transcript had itself been produced by a machine; nobody had ever corrected one by hand. The usual way to choose a speech model, scoring it against human transcripts, was not available.

Dialects are not equally hard. Maghrebi Arabic is the furthest from the standard language and the weakest spot for most speech models, and it was a large part of the archive.

The scale had to fit one machine. 62,205 recordings and about 21,300 hours of audio, some single sessions longer than seven hours, on one office computer that also restarts, loses power and has its archive disk unplugged.

Many of the voices are children’s. Sending the audio to a cloud transcription service would have meant another processor holding children’s voices. The work had to stay in the building.

What we built

Local by default. Transcription runs on one consumer graphics card in the office, with open-weight models loaded from local storage. No audio, and no text, goes to an outside service to be transcribed.

The model chosen by measurement, region by region. Two strong candidates were first run on 32 sessions, 4.6 hours spread across all five regions. Cohere’s Arabic model was closer to the existing text in every region and about six times faster than Whisper large-v3.

Because the existing transcripts were machine-made too, closeness to them proves little. So 146 sessions, 60 of them Maghrebi, were transcribed by three independent systems and compared word by word: where two agree against the third, the third is probably wrong. That majority vote gave each system an estimated error rate without any human reference.

Tuned on the hardest dialect. On the 59 Maghrebi sessions the chosen model at first trailed the earlier tool. Longer speech segments and a wider search cut the words it lost in disagreements by 21%, from 1,689 to 1,329, against 1,565 for the earlier tool. Across all five regions the same change cut them from 3,995 to 3,495. Grouping segments into batches then doubled the throughput in testing with no meaningful loss of accuracy.

Dialect routing tested, and deliberately not switched on. A dialect-identification model was tested as a router, to send each recording to a specialised model. It agreed with the country named in the session in only 78 of 140 cases, partly because the interviewer and the trainee often speak different dialects. Routing on a guess that is right just over half the time sends many recordings to the wrong model, so one model tuned on the hardest dialect runs for every region. Specialised Maghrebi models are shortlisted; one replaces it only if it beats it on the same fixed 59-session test set.

Every segment checked. Empty text over audible speech, looping repetition and implausible character rates are caught automatically. A flagged segment is redone once with a stronger setting and the better reading is kept; anything still flagged is counted per recording for a person to check. In the 202.7-hour pilot, 291 recordings ran with no failures, sessions up to 7.6 hours long ran at the same speed, 77 segments were fixed by the retry and 2 were left flagged.

Old and new compared, which found misfiled transcripts. Each new transcript is compared with the one already stored. About 9% of the stored transcripts in the pilot (25 of 281) turned out to belong to a different video. The split was clean, with no borderline cases, and 21 of the 25 were traced to the video they really describe: a whole batch had been shifted by a growing offset. Nothing was overwritten to fix it. Those videos are transcribed again and the old text stays where it is.

Built to be interrupted. The job starts with the machine, without anyone logging in. A watchdog restarts it after a crash or after 30 minutes without progress. It resumes from the last saved segment, even in the middle of a seven-hour recording, and waits for the archive disk instead of failing when it is unplugged.

A person has the last word. Each transcript is published to the catalogue with timed subtitles, stored as a separate version beside any older transcript. The team marks the version it trusts, and a correction becomes a new version rather than an edit.

The result

Measure Result
Recordings transcribed 62,205
Audio about 21,300 hours, five dialect regions
Speed in production about 32 times real time, around 770 hours of audio a day
Hardware one consumer graphics card in the office
Estimated word error, 146-session comparison 12.5%, against 14.0% for the earlier tool and 27.5% for the second candidate
Maghrebi words lost in disagreements 1,689 before tuning, 1,329 after
Segments still flagged after the automatic retry fewer than 1 in 10,000
Stored transcripts found to belong to another video about 9% of the pilot sample
Cloud transcription cost none

What we would tell another organisation

  • If no human-checked transcript exists, three independent systems voting tell you more than any single accuracy score.
  • Tune on your hardest case. The easy cases follow; the reverse is rarely true.
  • Do not route on a classifier you have not measured. A router that is right half the time is worse than one strong model.
  • Compare new output with old output. It finds data problems nobody was looking for.
  • Assume a long job will be interrupted, and make the interruption cost one segment, not one day.
Case studies

More of the work

Have a similar problem?

Tell us what the process looks like today. We will say whether this approach would work for you.