iFlytek Transcribe

The Problem

When processing meeting recordings, interviews, or course audio, the hard part is not only converting speech to text. Teams also need to handle task polling, quota selection and payment, downloading historical transcripts, and later summary work. This skill wraps the iFlytek transcription workflow into a consistent xftj.py CLI, so an agent can follow a fixed sequence instead of stitching together separate API calls.

How It Works

The skill is Python-based and exposes a small set of core commands:

• transcribe: uploads a new audio file, starts transcription, polls status, and returns available quotas;
• list / detail: retrieves historical transcription files and specific record metadata;
• download: fetches transcript text by hj_id, resolving originAudioId internally;
• ai-summary: retrieves or generates an AI meeting summary and polls until it is ready;
• validate / auth: checks whether the configured apiKey is valid and re-authenticates through a local OAuth callback when needed.

New-file transcription is a two-step flow: upload and start the task first, then let the user choose a quota before payment and download. For folders with many audio files, the skill recommends grouped upload, parallel polling, and batch aggregation to avoid long serial waits.

Boundaries and Caveats

This skill is suited to common audio formats such as mp3, wav, and m4a, with Chinese and English transcription support. Processing time scales with audio length, so polling should not be interrupted. Each transcription consumes account quota, so available credits should be checked before running large batches. AI summaries may take an extra 1-3 minutes; if polling times out, the agent can either continue waiting or fall back to generating a summary from the transcript text. The configuration lives in ~/.xftj/config.json, and the API key must be valid before execution.