Fliki's recent addition of Hindi voice support is a significant expansion for teams targeting the Indian market. The announcement highlights a large library of voices, but my initial concern is accuracy—specifically, the prosodic handling of English loanwords and technical terminology common in business and SaaS content.
I ran a preliminary test by generating a 90-second sample script. The script contained:
* Common business phrases like "customer relationship management (CRM)" and "return on investment (ROI)"
* A mix of Hindi and proper nouns (e.g., "Mumbai office," "Sharma ji")
* A sentence with a numerical figure ("quarterly growth of 12.5 percent")
The output was intelligible, but I noted a slight robotic cadence on the loanwords. The numerical reading was correct.
I'm looking for more comprehensive, real-world data before I can factor this into any vendor evaluation. Has anyone conducted more rigorous testing? Specifically:
* What is the word error rate (WER) on diverse, paragraph-length Hindi text?
* How does it handle domain-specific jargon (e.g., marketing, analytics, fintech terms)?
* Are there noticeable inconsistencies in voice quality across different "speaker" options?
* For automated video/podcast creation, how is the sync between the generated Hindi audio and on-screen visuals or subtitles?
Without quantified accuracy metrics, it's difficult to assess this feature's viability for professional, client-facing content.
- Mark
independent eye