We used Sembly for 6 months to transcribe our engineering standups and retrospectives. It worked, but the cost didn't scale for our team size (25 engineers). We built a simple internal alternative.
**Cost Breakdown (Monthly):**
* Sembly Pro: ~$300/month
* Internal Tool (OpenAI Whisper API + S3 storage): ~$35/month
**Our Implementation:**
We built a Slack bot that triggers a workflow on meeting end. It fetches the recording, processes it, and posts the transcript.
Key parts of the processor script:
```python
# Simplified core function
def transcribe_audio(audio_file):
client = OpenAI()
with open(audio_file, "rb") as file:
transcript = client.audio.transcriptions.create(
model="whisper-1",
file=file,
response_format="text"
)
return transcript
```
**Trade-offs:**
* **Lost:** Sembly's automatic chaptering and speaker-specific "voice" identification. We now use generic "Speaker 1, Speaker 2".
* **Gained:** Full control over storage/retention. Faster processing for our specific use case. No vendor lock-in.
For us, the 90% cost reduction was worth losing some polish. We only needed accurate text, not the extra features.
Benchmarks or bust.
Hey, thanks for sharing this breakdown! I'm on a small marketing team (8 people) and we've been using Sembly for about a year to transcribe our content planning syncs and client calls.
Here's my take on your trade-offs:
1. **Pricing and Scale**: You hit the exact pain point. Sembly's per-seat model (around $15/user/month for Teams) works for our size, but at your 25-engineer headcount, I've seen bills balloon to $300+. The API route is massively cheaper for bulk transcription once you have the volume.
2. **Speaker Diarization**: This is Sembly's real value-add for us. It gets speaker identities about 80% right after training, so our transcripts show "Laura" and "Dev" instead of "Speaker 2." The generic labels you have now would be a step back for our client call reviews.
3. **Deployment and Control**: Your internal tool wins on control. We're stuck with Sembly's 90-day retention unless we manually export. Your S3 setup is smarter for long-term logs. The Slack bot integration sounds similar to our setup, though.
4. **Setup and Maintenance Effort**: You absorbed the dev cost. For non-engineering teams like mine, the 2-hour Sembly setup (Slack + Google Calendar connect) versus a multi-week build is the deciding factor.
Given your team's size and that you only need accurate text, your internal tool is clearly the right call. If speaker names or automated summaries were critical, then Sembly would be worth revisiting.
Could you share more about how you handle timestamping or searching old transcripts in your system?