What is Call Transcription?
Call transcription, also called speech-to-text transcription, converts the audio from a voice or video call into written text. The process can transcribe calls either live - as a call or event happens - or use a recording of a past conversation.
For teams that manage customer calls, training programs, medical conversations, or legal records, call transcription helps turn spoken conversations into searchable text. As far more advanced search and analysis features exist for text than audio, many use cases need a text-based history of conversations. Additionally, real-time speech-to-text transcription services (such as Closed Captioning) increase accessibility which improves understanding for people who are hard-of-hearing or new to a language.
When it comes to voice calls, call transcription is often used in a business context, for example, to improve training and feedback for call center employees. Logging the context and words spoken in a call can help you identify business problems algorithmically, making it easier to deploy resources in an evidence-based manner. Additionally, call transcriptions and recordings are valuable for legal purposes, where contemporaneous transcriptions, recordings, and notes are superior to other types of records.
Twilio lets you add call transcription to Programmable Voice calls, such as sales or support calls:
- During a live call: Use Real-Time Transcriptions, which sends the call audio to AI-based speech-to-text engines, such as Google or Deepgram, and delivers the transcripts to your application through webhooks.
- After a call: Transcribe your call recordings with the Batch Transcription API.
- When a caller responds to a prompt: Use
<Gather>speech recognition to capture what the caller says.
Call transcription legality differs by jurisdiction. Some jurisdictions ban the transcribing or recording of calls or transcribing real-time speech over a call or video. Some require some or all parties in a conversation to provide informed consent. Twilio can't comment on the specifics of your local laws. Consult your relevant laws or consult with your legal representation for your unique situation.
Due to differences in volume, accents, timing, and connection quality, the final mixed track of a voice or video call can often be unintelligible even for professional human transcribers. So-called Single-Channel Recordings only store the one final mixed track pre-transcription. This can increase the eventual number of transcription errors, especially if participants are speaking at the same time.

The highest accuracy call transcription solutions record each (or all) sides of the call separately. With individual recordings, a Dual-Channel Recording solution (or Multi-Channel Recording solution) performs better at eliminating cross-talk and cancellation noise which would otherwise interfere with the final mix. It also prevents most (or all) misattribution errors.
To learn more about our dual-channel call transcription options, consult our blog post on the topic.
To transcribe a live call, start with the Real-Time Transcriptions reference. To transcribe a recorded call, follow the Batch Transcription quickstart. To record a call, use the Record TwiML Voice verb.
To analyze caller sentiment and other insights from your calls, use Conversation Intelligence.
Learn more about call transcription, speech recognition, and call recording with Twilio: