Anki Cards Builder

Build Anki-ready CSV files for vocabulary cards by adding pronunciation audio to rows of words.

The main script supports two modes:

• Full lookup mode: fetch definitions from OneLook when Back is empty, try Oxford pronunciation audio, and fall back to local TTS when Oxford audio is unavailable.
• TTS-only mode: generate local TTS audio for every input row and keep the existing Back explanation text.

Files

• anki_audio_fetcher_with_translations.py - main CSV processing pipeline.
• qwen_tts_fallback.py - local Qwen TTS helper used by the main script.
• r.sh - simple example runner for full lookup mode.
• audio/ - generated or downloaded audio files.

Input CSV

The input CSV must have a Front column containing the word or phrase to pronounce.

Recommended columns:

Front,Back,Audio,Definition,DL valid
entranceway,Passage providing entry into a place,,,

If optional columns are missing, the script creates them:

• Back - card back text. In TTS-only mode this should already contain your thesaurus explanation.
• Audio - path or source URL for the audio.
• Definition - Oxford definition URL in full lookup mode.
• DL valid - whether the audio file was successfully downloaded or generated.

When audio succeeds, the script appends an Anki sound tag to Back, for example:

Passage providing entry into a place [sound:entranceway__tts.ogg]

Usage

Full lookup mode:

.venv/bin/python anki_audio_fetcher_with_translations.py --audio_dir audio anki_audio_fetcher.csv anki_audio_fetcher_out.csv

TTS-only mode:

.venv/bin/python anki_audio_fetcher_with_translations.py --tts-only --audio_dir audio anki_audio_fetcher.csv anki_audio_fetcher_out.csv

TTS-only mode skips OneLook, Oxford lookup, definition URL checks, and remote audio downloads. It uses Front as the text to synthesize and preserves existing Back text.

Local TTS Requirements

The TTS fallback is configured in anki_audio_fetcher_with_translations.py:

LOCAL_TTS_FALLBACK_MODEL = "/Users/alekseiplotnitskii/.lmstudio/models/mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"
LOCAL_TTS_RUNTIME_PYTHON = ".venv-tts/bin/python"
LOCAL_TTS_OUTPUT_FORMAT = "ogg"

Required for TTS:

• .venv-tts with mlx_audio and huggingface_hub installed.
• The Qwen3 TTS model at the configured path.
• ffmpeg on PATH when output format is ogg.

The helper generates WAV internally and converts it to OGG for Anki.

Notes

• Re-running the script will not duplicate an identical [sound:...] tag.
• Generated audio, virtual environments, logs, and the default output CSV are ignored by git.
• Full lookup mode uses Selenium and ChromeDriver for OneLook scraping, plus HTTP requests for Oxford audio and definition pages.