Transcript, speaker diarization, voiceprint, SRT… turning a recording into notes involves a handful of terms that sound alike but mean different things. This page defines the 6 most common ones as briefly as possible, and explains which AIMochi feature each one maps to.
A transcript is a written record of spoken audio, written out sentence by sentence as it was actually said — the goal is to preserve the speaker's own words faithfully, not to rewrite or condense them. That makes it different from meeting minutes: minutes are an edited summary that keeps only decisions and action items, whereas a transcript keeps the full flow of the conversation, usually with timecodes and speaker labels, as raw material for whatever you write next. Interview research, legal and medical records, and subtitling all depend on transcripts.
In AIMochi: once you send a recording, the transcript is the first thing the system produces; summaries and polished write-ups are built on top of it. For the full workflow, see How to create a meeting transcript: from recording to finished minutes (in Chinese).
Speech-to-text (STT) is the technology that automatically recognizes a speech signal and converts it into the corresponding written text; it is also commonly called automatic speech recognition (ASR). Modern STT systems use deep learning models that map audio waveforms to text directly, and recognition quality depends on how clear the recording is, how much background noise there is, the speaker's accent, and how densely the audio is packed with specialist terminology. STT is the shared foundation beneath transcripts, live captions and voice assistants.
In AIMochi: speech-to-text supports more than 99 languages, covering Chinese, English, Japanese, Korean and Southeast Asian languages. For supported audio formats and upload methods, see the FAQ.
Speaker diarization is the technology that determines who spoke when in a multi-person recording, splitting the audio into segments and assigning each one to a different speaker. It answers "who was talking at this moment" rather than "what was said" — the latter is the job of speech-to-text, and you only get a transcript with speaker labels when the two work together. Diarization does not require knowing anyone's identity in advance, so its output is normally anonymous labels such as Speaker A and Speaker B.
In AIMochi: conversations are labelled automatically as Speaker A, B, C and so on, for up to 10 speakers. For practical tips on improving accuracy, see How to create a meeting transcript: from recording to finished minutes (in Chinese).
Voiceprint recognition identifies who a speaker is based on the individual acoustic characteristics of their voice, making it a form of biometric identification. The system extracts a numerical vector representing a person's vocal timbre (the voiceprint embedding), then compares two recordings to see whether their vectors are similar enough to be the same person. The difference from speaker diarization is one of purpose: voiceprint recognition establishes "who this is", while diarization only needs to work out "whether these segments came from the same person" — and it does so precisely by clustering on voiceprint similarity.
In AIMochi: voiceprint technology is what separates the different speakers in a recording, and it is why speaker labels can be generated automatically.
AI summarization is the process by which a language model automatically reads a longer piece of text and produces a substantially shorter version that keeps the key information. There are two broad approaches: extractive summarization picks out the most representative sentences from the original, while abstractive summarization reorganizes the content in its own words — which is what today's large language models mostly do. Applied to meetings, summaries are usually structured further into decisions, action items and owners.
In AIMochi: once the transcript is ready you can apply summary generation, note polishing, multilingual translation and smart Q&A directly. Each feature is explained in the tutorial.
An SRT subtitle file is a plain-text subtitle format with the .srt extension. Its contents are a series of subtitle blocks, each containing a sequence number, a start and end timecode, and the text to display. Timecodes are written as hours:minutes:seconds,milliseconds and joined by an arrow, and blocks are separated by a blank line. Because the format is so simple, virtually every player and video platform supports it, which makes SRT the most universal external subtitle format in use today.
In AIMochi: when exporting a voice note you can choose one sentence per line, merged paragraphs, or a subtitle file — and that subtitle file can be dropped straight onto a video.