Whisper (transcription)

What it does in ClientPulse: Picks the engine that turns meeting audio (from Zoom recordings, uploaded calls, etc.) into text. You choose between cloud, local, or a hybrid of both.

Status: Available

Tier: Available on all plans

Per-agency: One transcription mode per workspace.

Whisper is not an "integration" with a third-party account — it's a configuration choice about WHERE the audio gets transcribed. Cloud is the default and works out of the box.

Before you start (prerequisites)

There are three modes. Pick the one that fits your data-handling preferences.

Cloud mode (the default)

• What it is: ClientPulse sends audio to OpenAI's Whisper API for transcription.
• Prerequisite: Nothing — works out of the box.
• Cost: Included in your ClientPulse subscription.
• Trade-off: Audio briefly leaves your infrastructure (passes through OpenAI). OpenAI's Whisper API terms say they don't train on it, but if your clients have strict data-residency requirements you may want Local mode instead.

Local mode

• What it is: You run a self-hosted faster-whisper server inside your own infrastructure. Audio never leaves.
• Prerequisite: A machine running faster-whisper, reachable from the ClientPulse servers (or, if self-hosting ClientPulse, on your internal network). Typical setup: a GPU instance (NVIDIA T4 or better) running the official Docker image.
• Cost: Whatever your infrastructure costs (compute + GPU). Free if you already have the hardware.
• Trade-off: You operate the server. Slower than cloud unless you have GPU.

Hybrid mode

• What it is: Tries Local first, falls back to Cloud automatically if your local server is unreachable.
• Prerequisite: Same as Local — you need a working faster-whisper endpoint.
• Cost: Same as Local; Cloud fallback is included.
• Trade-off: Best of both worlds, but you still need the local server.

Step-by-step setup

1. (Local / Hybrid only) Run faster-whisper

Skip this section if you're using Cloud mode.

1. On a GPU host (or CPU host for low volume), run:
   `

   docker run -d --gpus all -p 8000:8000 \

   ghcr.io/systran/faster-whisper-server:latest-cuda

   `
2. Verify it responds: curl http://YOUR_HOST:8000/v1/audio/transcriptions should return a 400 (because we sent no audio) — but the fact that it responds at all means the server is reachable. Screenshot to be addedterminal showing curl response from the local whisper endpoint
3. Note the URL — e.g. http://10.0.1.42:8000. You'll paste this into ClientPulse next.

2. In ClientPulse

1. Open https://clientpulse.helloaurora.ai/dashboard/integrations.
2. Find the Whisper (Transcription) card. Click Configure Whisper to expand. Screenshot to be addedIntegrations page with Whisper card expanded
3. Pick a mode: Cloud, Local, or Hybrid.
4. If you picked Local or Hybrid, paste your faster-whisper endpoint into the Local Whisper endpoint field (e.g. http://10.0.1.42:8000).
5. Click Save.

3. Verify it's working

• The card shows Currently: Cloud / Local / Hybrid mode.
• Trigger a transcription by playing back a Zoom recording in ClientPulse, or upload a short audio file via Client → Files → Upload audio. Text should appear within a minute or two (cloud) / a few minutes (local CPU).

What you can do with it once configured

• Auto-transcribe Zoom cloud recordings as they come in.
• Auto-transcribe uploaded meeting audio.
• Feed transcripts into Aurora summaries on the client timeline.

Common pitfalls

• Pitfall: Local mode set but transcripts never appear. Fix: Your faster-whisper endpoint isn't reachable from ClientPulse. Check firewall rules, then click Send test message equivalent — for now, upload a 30-second audio file and watch dashboard/observability for transcription job logs. If you don't see jobs running, email support@helloaurora.ai with the endpoint URL (and confirm the URL is reachable from your laptop first with curl).
• Pitfall: Hybrid mode "stuck" on Cloud — local is never used. Fix: The fallback logic only goes Cloud when Local fails. Open dashboard/observability and look for local_whisper_error — usually a DNS or port issue.
• Pitfall: Audio quality is poor and transcripts are nonsense. Fix: This is a Whisper limitation, not a ClientPulse one — neither cloud nor local can produce good text from a bad recording. Check the Zoom recording quality settings.

Restrictions

• ClientPulse never stores raw audio long-term in Cloud mode — only the resulting transcript is persisted. In Local mode you control everything.
• Switching modes only affects future transcriptions. Existing transcripts stay as they were.
• The local endpoint must be HTTP-accessible from ClientPulse's servers (or your own deployment, if self-hosting CP).

Need help?

Email support@helloaurora.ai — we'll walk you through it.