Skip to main content
Use case

AI voice transcription for sales teams.

Prospects record a 60-second voice note. HeySpeak transcribes it in seconds and writes a context-aware summary. EU-hosted by default, no call needed.

Try it free
No login required for your prospects

The short answer

A prospect opens your Magic Link, records a 60-second voice note in their browser, and hits send. Mistral Voxtral transcribes it on EU servers within seconds. Mistral Small turns the transcript into a one-line summary plus intent signals. You read the result in the dashboard before you decide whether a call is worth booking.
Under 10 sec
typical transcription time for a 60-second note
EU-hosted
Voxtral runs on Mistral AI infrastructure in Europe
7+ languages
auto-detected, including German, French, Spanish

Why sales teams stopped trusting generic transcription

Most transcription tools were built for meetings. You record a 45-minute call, the tool spits out a wall of text, and someone has to read it later to find the one thing that mattered. That works for retros. It does not work for early-funnel prospect feedback, where the question is small and the time pressure is real.

The other problem is jurisdiction. Sales teams selling into the EU have to answer the GDPR question every time they pick a vendor. Most call-recording stacks store and process audio in the US, and the legal review takes longer than the integration. Voxtral on EU infrastructure removes that conversation by default for the primary path.

Be honest about the limits: transcription quality depends on audio clarity. A prospect recording from a quiet office gets a near-perfect transcript. A prospect on a windy street with traffic noise gets a transcript with gaps. The model is good. It is not magic.

The workflow

What happens between the prospect tapping send and you reading the summary.

  1. 1

    Prospect records a 60-second voice note

    You send a Magic Link with one question. The prospect opens it on any phone or laptop, taps record, and speaks for up to sixty seconds. No app, no login, no account. The audio uploads to a private Cloudflare R2 bucket the moment they hit send.

  2. 2

    Voxtral transcribes within seconds

    Mistral Voxtral runs the transcription on EU-hosted servers. Most 60-second notes finish transcribing in under ten seconds. If a request fails, the system retries automatically up to three times before marking the response for manual review. All transcription stays on Mistral AI infrastructure.

  3. 3

    Mistral Small writes the summary

    Once the transcript exists, Mistral Small reads it against the question you asked and produces a one-line summary plus the intent signals worth flagging: pricing concern, competitor mention, decision timeline, blocker. The summary is context-aware, not a generic abstract.

  4. 4

    Sales rep reads the result in the dashboard

    You open the dashboard. Each response shows the summary, the full transcript, and the original audio behind a 1-hour signed URL. Scan ten responses in a few minutes. Listen to the two that surprised you. Reply or queue the next move.

The stack, in plain terms

No black box. Three named models, each doing one job.

  • Mistral Voxtral. Primary transcription engine. EU-hosted. Handles audio in seven or more European languages with automatic detection. All recordings go through Voxtral.
  • Retry logic. If Voxtral fails or times out, the system retries automatically up to three times before marking the response for review. All retries stay on Mistral AI infrastructure. No audio leaves the EU.
  • Mistral Small. Context-aware summaries. Reads the transcript next to the question you asked, then writes a one-line summary that picks up pricing concerns, competitor mentions, decision timelines, and blockers. Not a generic abstract.

Related pages

Where AI transcription fits in a broader async voice sales flow.

Use case

What to send after a deal goes cold

Use the same Magic Link to ask lost prospects one question. Read the transcript pattern across five deals.

Playbook

Customer discovery, twenty interviews, no calls

A six-step playbook for running real interviews async. Transcripts and summaries do the synthesis work.

Use case

Async user interviews

When voice notes replace the live call, and when they do not. Worked examples and limits.

Guide

Voice to text feedback, end to end

Longer pillar guide: the full async voice feedback workflow, from question design to acting on the transcript.

Common questions

How accurate is the transcription?
Voxtral handles clear voice notes from a phone with strong accuracy, including across most major European languages. Quality drops with heavy background noise, multiple overlapping speakers, or strong accents the model was not trained on. The transcript is shown alongside the audio, so you can always check a passage by ear if a word looks off.
Is the transcription EU-hosted?
Yes. Mistral Voxtral runs on EU infrastructure, which matters for sales teams under GDPR or doing business with European prospects. All transcription and summarization routes through Mistral AI, 100% EU-hosted. No audio is ever sent to US providers.
What languages does it support?
Voxtral covers most major European languages out of the box, including English, German, French, Spanish, Italian, Portuguese, and Dutch. The summary model handles the same set. Detection is automatic, so you do not need to tell the system what language the prospect spoke.
How is this different from Gong, Fathom, or Otter?
Those tools transcribe live calls you already had. HeySpeak replaces the call. The prospect records a 60-second voice note instead of joining a Zoom, and you get a transcript and intent summary without anyone scheduling a meeting. Different layer of the funnel: Gong sits on top of calls, HeySpeak sits one step earlier and decides which calls are worth booking.
Can I export transcripts to my CRM?
Not yet via a native integration. For now, the workflow is read in the dashboard, copy the summary or transcript, paste into HubSpot or Pipedrive. A direct CRM sync is on the roadmap and the priority depends on how many sales teams ask for it. If you want it, tell us in the feedback link inside the dashboard.
How long are recordings stored?
Audio sits in a private Cloudflare R2 bucket and is only accessible through signed URLs that expire after one hour. The transcript and summary stay in your dashboard as long as your account is active. You can delete any response at any time from the dashboard, which removes the audio file and the transcript.

Stop transcribing calls. Skip the call.

Five free responses to start. Setup takes under a minute.

Create your first Magic Link