Assembly AI Best
Neural Network
Specifications
- Input
- audio
- Output
- text
Prompt cost
(per 1M tokens)
import OpenAI from 'openai';
const openai = new OpenAI({
apiKey: '<your bothub access token>',
baseURL: 'https://openai.bothub.chat/v1'
});
import fs from 'fs';
async function main() {
const { text } = await openai.audio.transcriptions.create({
file: fs.createReadStream('/path/to/file.mp3'),
model: 'assembly-ai-best',
});
console.log(text);
}main();
How it works Assembly AI Best?
Frequently asked questions about Assembly AI Best
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
assemblyai-best converts speech to text: podcasts, interviews, calls, lectures, voice messages. The large file processing mode allows you to upload long recordings in their entirety without cutting them manually. It is suitable for transcriptions, meeting notes, preparing subtitles, and searching through audio archives. Accessible from Russia without a VPN or foreign card.
The output limit is 4096 tokens, approximately several pages of text, and the context window is 4095 tokens. If the recording is long and the transcription does not fit into one response, split the audio into fragments and assemble the final text from the parts.
A separate reasoning mode is not noted in our data, and it is not needed for transcription: the model recognizes speech and returns text. If you need analysis of the transcription, a summary, or conclusions, pass the finished text to a text model — they are available in the same window in BotHub.
Support for function calling and strict JSON output is not noted in our data — rely on the standard text recognition result. The model itself can be called via the unified OpenAI-compatible BotHub API, and it is more convenient to structure the transcription by fields on your application's side.
Audio is the primary input: upload a recording, and the model will return text, including in large file mode. Images and video inputs are not noted in our data, so for video clips, it makes sense to first extract the audio track and send that.
The list of supported languages is not specified in our data, so we will not promise accuracy in Russian — check it with your own recording. The quality of the transcription depends most heavily on the clarity of the sound: a good microphone, minimal background noise, and no overlapping voices yield a noticeably better result.