iOS 18 Notes Transcription for Multiple Speakers

Apple's iOS 18 introduces live audio transcription directly inside the Notes app, turning spoken words into text in real time. This article breaks down how the feature handles multi-speaker scenarios, its ability to process overlapping speech, the limitations of on-device processing regarding speaker identification, and the best practices for getting clean transcripts in group settings.

Lack of Speaker Diarization

The live audio transcription feature in iOS 18 does not include speaker diarization. Speaker diarization is the process of partitioning an audio stream into distinct segments categorized by individual speaker identity (such as labeling lines as "Speaker 1" and "Speaker 2"). When recording a meeting or interview using Notes, the transcript generates a single, continuous block of text without attributing specific statements to different participants.

Behavior During Simultaneous Speech

When two or more individuals speak simultaneously, the transcription engine struggles to separate the overlapping acoustic signals. The on-device neural engine attempts to map all incoming sound to known language models, which generally leads to one of three outcomes:

  1. Volume Prioritization: The speech recognition model typically latches onto the louder or clearer voice, completely dropping the words of the secondary speaker.
  2. Interleaved Text: If both voices are at a similar volume, the model may attempt to transcribe phonemes from both sources at the same time, producing fragmented and nonsensical sentences that interweave words from both individuals.
  3. Dropped Words: Rapid crosstalk often causes the model to miss full phrases, leaving gaps in the resulting transcript.

Technical Limitations of On-Device Processing

Apple designed live transcription in Notes to run primarily on-device for speed, privacy, and offline accessibility. Multi-speaker voice separation and diarization require significant computational power and complex spatial or acoustic modeling. Because the iPhone processes this speech data in real time alongside Apple Intelligence tasks, the system prioritizes low-latency transcription of the primary audio stream over complex voice isolation.

Using Apple Intelligence Summaries to Clarify Transcripts

While the raw transcript does not separate speakers, Apple Intelligence provides an integrated summary feature for recorded notes. After the recording ends, users can generate a summary of the transcript. The large language model analyzes the context of the conversation and can often infer that multiple perspectives or action items exist, reorganizing messy crosstalk into a coherent, structured overview.

Best Practices for Multi-Speaker Recording in Notes

To obtain the most accurate transcription when multiple people are present, follow these operational steps:

  • Enforce Turn-Taking: Minimize simultaneous speaking. Pausing between speakers allows the model to process phrases cleanly and apply proper punctuation.
  • Microphone Placement: Position the iPhone equidistant from all participants. If one speaker is significantly closer to the microphone, their voice will overpower others during overlapping speech.
  • Reduce Background Noise: Eliminate ambient sound, as background chatter combined with multiple primary speakers degrades the transcription accuracy significantly faster than clean audio.