Use cases
For phone systems and IVR
The one place where u-law and a boring voice are the correct answers.
Last checked 2026-08-14
In short
Phone systems need short prompts, in telephony format, re-recordable whenever something changes. Naadly renders directly to 8 kHz u-law as well as MP3 and WAV, does bulk prompt sets from a CSV, and keeps one voice consistent across a menu tree - including in 23 Indian languages, which is what a support line in India actually needs.
Getting the format right
Telephony is 8 kHz u-law. Rendering to MP3 and converting works and loses quality twice; rendering to u-law directly does not. Prompts should be short, unhurried and free of music - a caller is listening for the number they need, not enjoying the read.
Writing prompts that work on a phone line
An 8 kHz line throws away most of the information a voice carries, so what survives is diction and pace rather than warmth. Write for the ear of someone in a car:
- Action before number - "for billing, press two", not "press two for billing" - because the caller is deciding while listening.
- Four options at most per menu, and no nested sentence.
- Digits spelled as words, and phone numbers grouped with commas so the reader pauses where a person would.
- A named fallback: "stay on the line and someone will answer".
- Preview on your longest prompt, not your shortest - that is where a fast voice becomes unintelligible.
Multilingual lines
A support line in India that answers in Hindi, Tamil and English needs three prompt sets with identical ids and one voice per language. Same CSV, same job, three passes. Keep the ids identical across languages: the PBX then selects a language by choosing a directory rather than by a second mapping nobody maintains.
Questions
- Can you output u-law for a PBX?
- Yes - 8 kHz u-law is a first-class output format alongside MP3, WAV, OGG and FLAC.
- Can I regenerate prompts automatically?
- Yes, through the API - most people wire the prompt CSV into their own deployment.
- Which voice suits IVR?
- A deliberate, neutral, clearly-graded voice in the caller's accent. Filter by pace and accent, and preview on your longest prompt.