Transcription & AI

Transcription is one of the most essential skills for a linguist who is studying new languages. Languages have a wide variety of scripts and some languages might not have a written script all together. Therefore, linguists needed a standardized way to be able to document languages. To solve this, they developed the IPA. I talk more about the IPA in my previous post, you can check it out here if you want. Essentially, the IPA has a symbol for every sound, allowing linguists to document things like accents in an easier manner than using traditional scripts. Transcription is the process of converting audio recordings of people speaking a language into IPA. This is a labor-intensive process, and also a somewhat subjective one. Even though IPA is supposed to have a symbol for every sound, it can be difficult to always pinpoint which sounds are being heard (take for example h vs ɦ).

However, with advancements in AI, I wonder if transcription will still be done by hand. In theory, it should be easier, not harder, for AI to transcribe sound into IPA rather than normal voice recognition. Currently, the main problem with AI is it often fails to recognize certain accents, especially those that are not spoken as frequently. However with IPA, that shouldn’t be a problem as each accent would use different characters. I wonder if there is already AI being used for IPA translation and if transcription will still be a skill that people need to learn in the future.