Custom phonetic pronunciations in AI transcription

Request
Enable admins to define custom phonetic pronunciations for proper nouns within AI transcription (e.g. agent names, customer names, product names), improving transcription accuracy in call summaries.

What does this solve?
The AI-generated call summaries for Talk conversations are valuable, but transcription accuracy breaks down for non-Anglophone names (e.g. Polish, Zulu, Xhosa). Names are frequently mis-transcribed, which:

• Reduces trust in AI-generated summaries
• Requires manual corrections by agents or QA teams
• Creates an inconsistent experience for non-Anglophone staff
• Impacts downstream reporting and record accuracy

Existing functionality such as “boost keywords in transcriptions” does not address this, as it influences spelling recognition rather than pronunciation at the speech-to-text level.

When did this issue last affect you?
Daily—our teams handle calls continuously, and transcription errors occur regularly.

What would your ideal situation look like?
A configurable phonetic dictionary that allows admins to define how specific terms should be recognised in speech.
For example:

• Written name: Xhantilomzi
• Phonetic input: KAN-TEE-LOM-ZEE (or IPA support)

Optional enhancements:

• Support for Internation Phonetic Alphabet (IPA) notation
• Ability to upload audio samples for pronunciation learning
• Application across agent names, customer names, and other proper nouns

Why this matters
As AI transcription becomes a core part of CX workflows, it must perform reliably across global, diverse teams. Supporting phonetic coaching improves:

• Accuracy of AI outputs
• Agent experience and inclusivity
• Operational efficiency
• Trust in automation