Fish Audio: Transcribe 1 Pro

Fish Audio: Transcribe 1 Pro is a speech-to-text model from Fish Audio, a provider known for its work in audio and speech processing. It sits in the transcription category and is positioned as the company's pro-tier transcription offering. The provider released it on 2026-09-24, so it is a recent addition to the Fish Audio line rather than a long-standing legacy model.

The model takes audio as its only input and returns a transcription as its output. It is tuned specifically for interviews, meetings, and podcasts, which means it is built for long-form, multi-speaker recordings rather than short voice commands. Its distinguishing feature is inline speaker markers, so the transcript labels who is speaking rather than producing one undifferentiated block of text. It also preserves emotion and vocal-event cues, meaning non-lexical information in the audio is carried into the transcript instead of being stripped out. No context window length is published for this model, so plan around the size of the audio you send rather than a stated token budget.

Those capabilities point to a fairly specific set of uses. Diarized transcripts are useful for meeting notes where each attendee's remarks need to be attributed, for interview transcripts that a journalist or researcher will quote from, and for podcast production where show notes, chapters, or searchable transcripts depend on knowing who said what. Because emotion and vocal-event cues survive transcription, the output can also support downstream tasks like sentiment analysis on call recordings, editorial review of interview tone, or accessibility captions that convey more than bare words. Any pipeline that currently pairs a plain speech-to-text model with a separate diarization step is a candidate, since this model folds speaker labeling into the transcription itself.

You call it through api14's OpenAI-compatible API using the model id "fish-audio/transcribe-1-pro", so existing OpenAI SDK clients and HTTP integrations work with a change of base URL and model name. Requests follow the standard audio transcription shape: send the audio file, and read the transcript from the response. The model exposes no additional request parameters beyond that, so there is nothing to tune for temperature, prompt style, or output format. Treat it as a fixed configuration: what you send is audio, and what you get back is a transcript with speaker and emotion information embedded.

Because the provider lists no supported parameters and no maximum output length, this model is a poor fit if you need control over decoding or a specific structured output schema from the API itself. If you want adjustable parameters, a plain text model, or audio input combined with reasoning over the content, another model on api14 will serve you better. Choose Transcribe 1 Pro when clean, speaker-labeled transcription of interviews, meetings, or podcasts is the whole job.