Setting up call transcription in Bitrix24

Setting Up Call Transcription in Bitrix24

Everyone records calls — but no one is going to manually listen through hours of manager conversations. This is where transcription converts audio into text that can be searched, analyzed, and passed into CRM without ever pressing play.

How Transcription Works in Bitrix24

Bitrix24 uses the built-in Bitrix24 Contact Center module together with a cloud speech recognition service. After a call ends, the recording is sent via the telephony module to the recognition server — the result is returned as text and saved to the CRM entity card (lead, deal, contact).

The technical process looks like this:

1. The call ends; the recording is saved in b_voip_call linked to a CALL_ID.
2. A transcription task is placed in the processing queue.
3. The Speech-to-Text service (Yandex SpeechKit or the built-in Bitrix engine) receives the audio file.
4. The text is returned and written to the TRANSCRIPT field of the call record.
5. A "Call Transcript" block appears in the CRM card, split by speaker.

What Is Required to Enable It

Transcription is available only on the Professional and Enterprise plans (Bitrix24 cloud). For the on-premise version — via an external Speech-to-Text provider and the REST API method telephony.externalcall.finish.

Activation steps:

1. CRM → Telephony → Settings → "Transcription" tab.
2. Enable the "Transcribe calls" toggle.
3. Select the recognition language (Russian, English — the list depends on the plan).
4. Specify which numbers to enable transcription for.
5. Configure access rights: who can see transcripts (manager, the agent themselves, everyone).

After activation, every new call longer than 10 seconds will be transcribed automatically. Older recordings — only via manual trigger or a REST script.

Limitations and Common Issues

Recognition quality depends heavily on recording quality. If there is noise, echo, or a low bitrate on the line — accuracy drops to 60–70%. Standard bitrate for acceptable recognition: 8 kHz mono minimum, 16 kHz preferred.

Speaker separation (diarization) only works with stereo recordings where the agent and client are recorded on separate channels. With mono recordings, transcription produces a single undifferentiated text with no "who said what" markup.

Quotas. Each plan has a monthly transcription minute limit. When exceeded, transcription stops until the next billing period. Limits are visible under "Plans" → "Usage".

On-premise version. The built-in engine is unavailable. Integration with Yandex SpeechKit or Google Speech-to-Text is implemented via the OnVoipCallEnd event handler, calling the external API and writing the result back to Bitrix via CRest::call('telephony.externalcall.finish', [...]).

What to Do with the Text After Transcription

Transcripts can be used in robots and business processes: trigger "Call ended" → condition on transcript text → action (stage change, task to manager, notification). The transcription field is available as the variable {=Document.TRANSCRIPT} in the business process designer.

Searching through transcript texts works via the standard CRM search — the transcript is indexed in b_search_content alongside other fields.