Transcription

What is automated transcription?

Automated transcription is the use of speech recognition to convert spoken audio, radio communications included, into written text without a human typing it. Audera records a live source continuously during the monitoring period, divides the recording into short segments and transcribes each segment. The text is stored with the time it was spoken, its duration and a reference to the source audio file. The result is a timestamped transcript of the entire monitoring period.

How speech becomes searchable text

1. Recording. The feed is captured for the whole monitoring period and stored in time-ordered segments.
2. Segmentation. Long recordings are split into short pieces so that each transmission can be located precisely.
3. Speech recognition. Each segment is transcribed. Silence and dead air are detected and skipped so the transcript contains only communications.
4. Timestamping. Each transcript line receives an absolute timestamp and a reference to the file and offset it was produced from.
5. Indexing. The transcript is organized so it can be searched by word, phrase, time range or detected entity.

What a timestamped transcript looks like

Transcripts are delivered as structured data and as readable text. The illustrative extract below shows the shape of the output. It is an example of the format, not a record from a real customer job.

Each line can be searched, filtered and traced back to the recording. When keyword tracking or event detection is part of the job, the same timestamps are used to build the event timeline. Read about audio intelligence and event detection.

Transcription accuracy on radio and other difficult audio

Radio audio is more difficult to transcribe than studio speech. Narrow bandwidth, background noise, overlapping transmissions, abbreviations, callsigns and local terminology all affect accuracy. Audera's transcripts are designed to be read together with the recording rather than as a replacement for it: every line references the audio it came from, and the audio archive is always part of the delivery. Where you provide domain vocabulary such as unit names, locations or codes in advance, it is used to improve keyword matching and review.

Transcription as part of a monitoring job

Transcription is included in every monitoring job. If you already hold recordings, Audera can also transcribe and analyze them without live monitoring. In both cases the deliverable is the same: a searchable, timestamped transcript, the original audio and, where requested, detected events and a report.