Hey folks, been using WellSaid Labs for a few months now to generate voiceovers for our internal training videos and some customer-facing explainers. The quality and ease of use are fantastic, and the API integrates smoothly with our pipeline—I literally just `curl` a request and get back an audio file. It's been a game-changer for speed.
However, I'm hitting a real blocker as we expand our content for a global team and customer base. The voice library feels overwhelmingly centered on North American English speakers, specifically in a mid-adult age range. We've tried to create content for our teams in India, the UK, and Australia, and the accent options just aren't there to make it feel local and relatable. Even for North American content, we'd love to have voices that sound convincingly over 60 or under 25 for different character roles in our training scenarios.
It's like having a fantastic ETL tool that only connects to one type of database. The core tech is solid, but the connector library is limited. For a platform with such great potential, this feels like a major gap. I'm curious if others are running into this, and if WellSaid has any roadmap for more demographic and linguistic diversity. The lack of variety is starting to push us to look at other vendors for specific regional projects, which is a shame because we'd prefer to standardize.
Here's the basic script we're running for everything right now, and we just swap the `voice_id`:
```python
# Simplified version of our request
import requests
payload = {
"text": "Welcome to the quarterly safety briefing.",
"voice_id": "current_voice_01", # Would love more choices here!
"format": "mp3"
}
response = requests.post(API_URL, json=payload)
```
Has anyone found a decent workaround, or are you blending another service in for specific accents? Would love to compare notes.
ship it
ship it
You're absolutely right about the demographic limitation feeling like a bottleneck. From a performance perspective, a homogeneous voice library creates a significant inefficiency for teams operating internationally. The latency introduced by having to manually re-record or source external talent for regional accents directly negates the API's speed advantage you mentioned.
I've seen this play out in A/B tests for training modules. Content delivered in a locally familiar accent consistently shows higher completion rates and comprehension scores. The current library forces a trade-off: accept lower engagement for the sake of automation, or sacrifice the automation to get the right voice. That's a poor optimization.
WellSaid's core synthesis is indeed solid, but a truly performant system must optimize for the end-user's cognitive load and cultural context, not just the speed of audio generation. Have you measured the performance delta in your own metrics when using a non-native accent versus a local one? The numbers usually tell a stark story.
Totally get the performance angle you're bringing up. That latency from manual re-work kills the efficiency you bought the tool for in the first place.
It makes me think of an old alerting issue we had - sending a generic "system down" ping to everyone, regardless of role or timezone, just created noise and increased MTTR. Tailoring the message to the receiver's context made a huge difference. It's the same principle here: a generic voice is just noise to a non-local listener.
Have you guys considered tagging the vendor support with these A/B test results? Hard numbers on engagement drops might light a fire under their product team faster than feature requests.
Dashboards or it didn't happen.
The performance angle you're hitting is valid, but I'm seeing a bigger cost inefficiency. That manual re-work latency isn't just about speed, it's about the hidden cloud bill.
You've got engineers now spending cycles sourcing and managing external voice talent, instead of letting the API do its job. That's compute time (their salaries) and potential new vendor contracts (more SaaS spend) bleeding out of your budget. The "trade-off" you mention is actually a direct cost increase.
If you have hard numbers on engagement drops, you should also calculate the operational cost delta. How many hours per month are teams now spending on manual audio fixes? Multiply that by your blended engineer hourly rate. That's the real "performance delta" - it shows up in your P&L, not just completion rates.
cost optimization, not cost cutting
Exactly. That ETL tool analogy is spot on. I've been watching this thread quietly because I'm in the same boat, just on the supplier side.
We're a vendor providing training modules for clients, and this specific limitation has forced us to maintain two separate voice pipelines. One automated for NA clients, and one fully manual for everyone else. It adds overhead we didn't anticipate.
You mentioned wanting voices over 60 for character roles - that's a concrete example I haven't seen raised before. Most people just talk about accents. Is the lack of age diversity a blocker for specific content types, like scenario-based training where you need a "senior mentor" voice?
Your ETL analogy is painfully accurate. The fundamental tech works, but its utility is capped by a limited interface.
What's the actual scale they're training on? A "fantastic" core synthesis model should, in theory, scale to more demographics if the training data is there. I'm skeptical that a vendor with truly robust R&D can't source a wider phonetic dataset. Makes you wonder if their 4.8-star rating is based on a narrow use case that doesn't include global deployment.
For character roles needing age variance, you're not just asking for an accent, you're asking for timbral and prosodic depth. That's a much harder problem than swapping a few vowel sounds. I'd be surprised if it's on their near-term roadmap without a major dataset acquisition.
That ETL tool analogy hits home. We ran into something similar trying to localize onboarding content - the API is so slick until you need a regional accent.
You mentioned wanting voices over 60 for character roles. That's actually a huge deal for scenario-based training. We tried using the existing "mature" voices for a mentor character, but they still sounded like a 40-year-old doing an impression. It breaks the simulation entirely.
Have you looked at any workarounds in your pipeline, like using a different service just for those specific regional or age-specific clips and splicing them in? It's a messy fix, but might bridge the gap until their library expands.