Real-Time Transcription
A Real-Time Transcription is a Service Cloud Voice feature that converts the audio of a live phone call into text as the conversation happens.
Definition
A Real-Time Transcription is a Service Cloud Voice feature that converts the audio of a live phone call into text as the conversation happens. The text streams onto the agent's Voice Call record during the call, so the agent reads what was said without taking notes by hand.
The transcript is more than a convenience for the agent. It is the data feed that lets Einstein analyze the call live, surface knowledge articles, suggest next actions, and gauge sentiment. Supervisors can watch the same transcript to triage a call in progress and step in when a rep needs help.
How Service Cloud Voice turns live audio into text
What real-time transcription actually does
Real-time transcription listens to a connected voice call and writes each spoken turn to text within seconds. Salesforce supports real-time transcription, not a post-call-only pass, so the words land on the Voice Call record while the customer is still talking. Each line is labeled by speaker, which keeps the agent line separate from the customer line as the dialog scrolls. The transcript appears in the Enhanced Conversation component on the Voice Call record page. An agent reads it the way they would read a chat thread, scrolling back to confirm a number or a name without asking the customer to repeat themselves. After the call, the same transcript stays attached to the record for review, coaching, or compliance. Behind the scenes, the audio is streamed to a speech-to-text service, the returned text is written back to Salesforce, and the component refreshes to show new lines. Because the text exists as structured data on the record, other features can read it. That is what makes transcription the entry point for Einstein on live calls, rather than a standalone note-taking tool.
Where it fits in the three Voice telephony models
Service Cloud Voice ships in three telephony models, and transcription is available across them with different plumbing. In Service Cloud Voice with Amazon Connect, Salesforce provides the contact center and you wire transcription through Amazon. In Service Cloud Voice with Partner Telephony, a certified partner such as a contact center vendor supplies the calls, and the partner sends transcript text into Salesforce. The Bring Your Own Telephony model lets you connect an existing Amazon Connect instance you already run. The model you pick decides who is responsible for the speech-to-text engine. With Amazon Connect, the engine is Amazon Transcribe or Contact Lens for Amazon Connect. With partner telephony, the provider runs its own engine and pushes the results in. The Help docs are explicit that for a non-Amazon partner you should contact the telephony provider to confirm transcription support and get setup details. No matter the model, the agent-facing result is the same. Text shows up in the Enhanced Conversation component in near real time, and the finished transcript is saved to the Voice Call record once the call ends.
The 15-minute limit and why Contact Lens matters
There is a real limitation worth knowing before you promise full-call transcripts. If your Amazon Connect contact flow uses Amazon Transcribe on its own, only the first 15 minutes of a call are transcribed. The cause is a Lambda function limit in that transcription path, not a Salesforce setting you can raise. For a quick support call this is fine, but a 40-minute troubleshooting session would lose most of its text. The fix is Contact Lens for Amazon Connect. When you enable Contact Lens and real-time transcription inside your Amazon Connect contact flow, calls transcribe past the 15-minute mark. Salesforce documents this as the way to transcribe longer calls, so teams that handle long conversations should plan on Contact Lens rather than the bare Amazon Transcribe path. This choice has knock-on effects. Contact Lens also produces its own analytics in Amazon, and it changes how minutes are metered. Treat the transcription engine as an early design decision, because switching it later means reworking the contact flow and re-validating that the Enhanced Conversation component still receives text.
Why the transcript is the foundation for Einstein
The strongest reason to turn on real-time transcription is what it unlocks downstream. Once spoken words exist as text on the record, Einstein can read them live. Salesforce describes Einstein analyzing the conversation to serve up the right knowledge articles and next actions for reps as the call unfolds. Without a transcript, those live recommendations have nothing to read. This is why mature contact centers treat transcription as a prerequisite rather than a nice extra. Einstein Next Best Action can fire recommended offers or steps based on what the customer just said. Knowledge surfacing pushes a relevant article to the agent at the moment the topic comes up. Sentiment signals can flag a call that is turning negative so a supervisor steps in early. After the call, the transcript feeds Einstein Conversation Insights and AI-generated call summaries, which means less manual wrap-up. The quality of all of this depends on the quality of the transcript. A clean, accurate transcript gives Einstein good input. A garbled one degrades every feature that reads it, which is why audio quality and language settings deserve attention.
The agent and supervisor experience
For the agent, the value is focus. Instead of typing notes while listening, the rep watches the transcript build and spends attention on the customer. They can scroll back to re-read a policy number or an address, and the running text reduces the chance of mishearing. When the call ends, the transcript is already saved, so wrap-up starts from a written record rather than memory. For the supervisor, real-time transcription turns voice into something you can watch like a text channel. Salesforce notes that supervisors can monitor multiple call transcriptions at once, in real time, alongside other digital channels. That lets a lead scan several live calls, spot one going sideways, and offer a hint or take over without having to listen to each call audio stream one at a time. There is a coaching payoff too. Because transcripts and recordings persist, a supervisor can review past calls, mark good and bad moments, and use real examples in training. The transcript becomes a searchable record of what was said, which is far easier to coach from than re-listening to hours of audio.
Setup, permissions, and common gotchas
Getting transcription on screen takes work in two places. First, transcription has to be enabled in the telephony layer, which for Amazon Connect means configuring it in the contact flow and choosing Amazon Transcribe or Contact Lens. Second, an admin adds the Enhanced Conversation component to the Voice Call record page in Lightning App Builder, because without that component the text has nowhere to display even when it is being generated. Permissions and language settings matter. Agents need the right Service Cloud Voice permission set licenses to see and work calls, and the transcription language must match how customers actually speak. A mismatch produces a transcript full of nonsense that also poisons the Einstein features reading it. Plan for the limits. The 15-minute Amazon Transcribe cap is the classic surprise. Test a long call end to end before going live, watch that text keeps flowing past 15 minutes, and confirm the finished transcript attaches to the record. Validate in a sandbox or test queue first, since calls placed before everything is wired correctly will not be transcribed retroactively.
How to set up real-time transcription in Service Cloud Voice
Real-time transcription needs configuration in the telephony layer and a component on the page before agents see any text. These steps cover the Salesforce side for a Service Cloud Voice with Amazon Connect setup. Confirm the exact labels in your org, since they shift slightly by Voice model and release.
- Confirm Service Cloud Voice is set up
Make sure Service Cloud Voice is provisioned and your telephony model (Amazon Connect, partner telephony, or BYOT) is connected. Transcription rides on top of a working Voice setup, so the call integration must function first.
- Enable transcription in the contact flow
In your Amazon Connect contact flow, enable real-time transcription. For calls longer than 15 minutes, enable Contact Lens and real-time transcription rather than relying on Amazon Transcribe alone, which stops transcribing after 15 minutes.
- Add the Enhanced Conversation component
In Lightning App Builder, open the Voice Call record page and drag the Enhanced Conversation component onto the layout. This is where the live transcript renders for the agent, so save and activate the page for the right app and profiles.
- Assign permissions and set language
Grant agents the required Service Cloud Voice permission set licenses, and set the transcription language to match your customers. A wrong language setting yields an unusable transcript and breaks the Einstein features that read it.
- Test a real call end to end
Place a test call, confirm text appears in near real time, and let it run past 15 minutes to verify Contact Lens keeps transcribing. Check that the completed transcript stays attached to the Voice Call record after the call ends.
Amazon Transcribe, Contact Lens for Amazon Connect, or a partner telephony provider's own engine. Contact Lens is required to transcribe calls longer than 15 minutes.
The Voice Call record page component that displays the live and saved transcript. Without it, transcription runs but agents see nothing.
The spoken language the engine listens for. It must match the customer's language, and supported languages depend on the transcription service you chose.
Optional layers that consume the transcript, including knowledge surfacing, Einstein Next Best Action, sentiment, and post-call Conversation Insights and call summaries.
- Amazon Transcribe on its own transcribes only the first 15 minutes of a call because of a Lambda function limit. Use Contact Lens for longer calls.
- If the Enhanced Conversation component is not on the Voice Call record page, agents see no transcript even when transcription is enabled.
- Calls placed before transcription and permissions are fully configured are not transcribed after the fact, so test before going live.
- A transcription language that does not match the spoken language produces garbled text and degrades every Einstein feature that reads it.
Prefer this walkthrough as its own page? How to Real-Time Transcription in Salesforce, step by step
Trust & references
Cross-checked against the following references.
Straight from the source - Salesforce's reference material on Real-Time Transcription.
Hands-on resources to go deeper on Real-Time Transcription.
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He runs salesforcedictionary.com to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
Q1. What does Real-Time Transcription do in Salesforce Service Cloud Voice?
Q2. Which capability depends on Real-Time Transcription being active on a call?
Q3. Why is Real-Time Transcription valuable for contact-center quality and productivity?
Discussion
Loading discussion…