This is an English translation of a Japanese blog. Some content may not be fully translated.
📝

Auto-Generate English Phrase Cards with AnkiConnect and Google Cloud TTS

📖 2 min read
Anki Google Cloud TTS

Introduction

I use Anki to memorize English phrases, but manually registering cards and attaching audio is tedious enough to make me fall off the habit. By combining the REST API of AnkiConnect with the Neural2 voices of Google Cloud Text-to-Speech , you can automate everything from card registration to high-quality audio attachment using only curl commands. This article documents that process.

Here is an overview of the entire workflow.

flowchart LR A[Phrase list<br/>text file] --> B[AnkiConnect API<br/>Create deck & add notes] A --> C[Google Cloud TTS API<br/>Neural2 audio generation] C --> D[MP3 file] D --> E[AnkiConnect API<br/>Save media & update fields] B --> F[Anki<br/>Card with audio complete] E --> F

Prerequisites

Item Details
Anki Version 23.10.0 or later
AnkiConnect Plugin code 2055492159
gcloud CLI Authenticated
Google Cloud Text-to-Speech API enabled

To install AnkiConnect, open Anki and go to Tools → Add-ons → Get Add-ons, then enter the code 2055492159. A restart of Anki is required after installation. AnkiConnect starts an HTTP server on port 8765 and accepts JSON-formatted requests.

Enable the Google Cloud Text-to-Speech API with the following command.

gcloud services enable texttospeech.googleapis.com

Bulk-Register Cards via the AnkiConnect API

Check the connection

First, verify that AnkiConnect is running.

curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{"action": "version", "version": 6}'

{"result": 6, "error": null}

Create a deck

Use the createDeck action to create a new deck.

curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "action": "createDeck",
    "version": 6,
    "params": {"deck": "English Meeting Phrases 15"}
  }'

Check the model (note type)

Before adding cards, confirm the model name and its field names.

# List all models
curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{"action": "modelNames", "version": 6}'

# List fields for a model
curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "action": "modelFieldNames",
    "version": 6,
    "params": {"modelName": "Basic"}
  }'

Bulk-add notes

Use the addNotes action to register multiple notes at once. The example below adds English meeting phrases in a Japanese-prompt → English-answer format for instant translation practice.

curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "action": "addNotes",
    "version": 6,
    "params": {
      "notes": [
        {
          "deckName": "English Meeting Phrases 15",
          "modelName": "Basic",
          "fields": {
            "Front": "Ask someone to repeat themselves when you did not catch what they said",
            "Back": "Sorry, could you say that again?"
          },
          "tags": ["Week1", "clarification"]
        },
        {
          "deckName": "English Meeting Phrases 15",
          "modelName": "Basic",
          "fields": {
            "Front": "Ask someone to speak more slowly when they are talking too fast",
            "Back": "Could you slow down a bit?"
          },
          "tags": ["Week1", "clarification"]
        }
      ]
    }
  }'

The result array in the response contains the ID of each note. A null value indicates that a duplicate note was found and skipped.

{"result": [1772875633429, 1772875633437], "error": null}

If you use a custom model, specify fields to match that model’s fields. For example, if your model has fields like Example, Synonyms, and GrammarNote, you can populate all of them in the same request.

Generate Audio with Google Cloud TTS

Set the quota project

When using gcloud CLI user credentials, you must specify the quota project via the x-goog-user-project header. Without it, you will receive a PERMISSION_DENIED error.

TOKEN=$(gcloud auth print-access-token)
PROJECT=$(gcloud config get-value project 2>/dev/null)

Generate an MP3 with a Neural2 voice

Google Cloud TTS Neural2 voices produce more natural-sounding speech compared to Standard or WaveNet voices, making them well-suited for language learning.

curl -s -X POST \
  -H "Authorization: Bearer $TOKEN" \
  -H "x-goog-user-project: $PROJECT" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {"text": "Sorry, could you say that again? I missed the last part."},
    "voice": {"languageCode": "en-US", "name": "en-US-Neural2-J"},
    "audioConfig": {"audioEncoding": "MP3", "speakingRate": 0.9}
  }' \
  "https://texttospeech.googleapis.com/v1/text:synthesize" \
  | python3 -c "
import sys, json, base64
data = json.load(sys.stdin)
sys.stdout.buffer.write(base64.b64decode(data['audioContent']))
" > meeting_phrase_01.mp3

Key parameters are as follows.

Parameter Description Example value
voice.name Voice model en-US-Neural2-J (male), en-US-Neural2-F (female)
audioConfig.speakingRate Speaking rate (0.25–2.0) 0.9 (slightly slower, good for learning)
audioConfig.audioEncoding Output format MP3, OGG_OPUS, LINEAR16

The API response contains base64-encoded audio data in the audioContent field; decode it and write it to a file.

Free tier

According to the Google Cloud TTS pricing page , Neural2 voices are free up to 1 million bytes per month. Since English phrases consist of ASCII characters (roughly 1 character ≈ 1 byte), personal English study usage is very unlikely to exceed the free tier.

Attach Audio Files to Anki Cards

Store the media file

Use the storeMediaFile action to save an MP3 file to Anki’s media folder. Pass the file as a base64-encoded string.

B64=$(base64 -i meeting_phrase_01.mp3)

curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d "{
    \"action\": \"storeMediaFile\",
    \"version\": 6,
    \"params\": {
      \"filename\": \"meeting_phrase_01.mp3\",
      \"data\": \"$B64\"
    }
  }"

Update the note field

Retrieve the note ID with findNotes, then use updateNoteFields to set the Sound field to [sound:filename].

# Get note IDs
curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "action": "findNotes",
    "version": 6,
    "params": {"query": "deck:English Meeting Phrases 15"}
  }'

# Update the Sound field
curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d '{
    "action": "updateNoteFields",
    "version": 6,
    "params": {
      "note": {
        "id": 1772875633429,
        "fields": {
          "Sound": "[sound:meeting_phrase_01.mp3]"
        }
      }
    }
  }'

Note on card templates

If you store the value as [sound:filename.mp3] in the Sound field, writing [sound:{{Sound}}] in the card template will double-wrap it and audio will not play. Write only {{Sound}} in the template.

<!-- Bad: double-wrapping prevents playback -->
[sound:{{Sound}}]

<!-- Good: the field value is expanded as-is -->
{{Sound}}

Batch Processing Script

Below is an example shell script that combines all of the steps above. It reads from a phrase list file and runs the full pipeline — card registration → TTS generation → audio attachment — in a single pass.

#!/bin/bash
set -euo pipefail

DECK="English Meeting Phrases 15"
MODEL="Basic"
VOICE="en-US-Neural2-J"
RATE=0.9
TOKEN=$(gcloud auth print-access-token)
PROJECT=$(gcloud config get-value project 2>/dev/null)

# Create deck
curl -s http://localhost:8765 -X POST \
  -H "Content-Type: application/json" \
  -d "{\"action\": \"createDeck\", \"version\": 6, \"params\": {\"deck\": \"$DECK\"}}" > /dev/null

# Phrase definitions (TSV: prompt<TAB>English phrase<TAB>example sentence<TAB>tag)
while IFS=$'\t' read -r front back example tag; do
  # Add note
  escaped_back=$(printf '%s' "$back" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))")
  escaped_front=$(printf '%s' "$front" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))")

  note_id=$(curl -s http://localhost:8765 -X POST \
    -H "Content-Type: application/json" \
    -d "{
      \"action\": \"addNote\",
      \"version\": 6,
      \"params\": {
        \"note\": {
          \"deckName\": \"$DECK\",
          \"modelName\": \"$MODEL\",
          \"fields\": {\"Front\": $escaped_front, \"Back\": $escaped_back},
          \"tags\": [\"$tag\"]
        }
      }
    }" | python3 -c "import sys,json; print(json.load(sys.stdin)['result'])")

  echo "Added note: $note_id - $front"

  # Generate TTS
  escaped_example=$(printf '%s' "$example" | python3 -c "import sys,json; print(json.dumps(sys.stdin.read()))")
  fname="phrase_${note_id}.mp3"

  curl -s -X POST \
    -H "Authorization: Bearer $TOKEN" \
    -H "x-goog-user-project: $PROJECT" \
    -H "Content-Type: application/json" \
    -d "{
      \"input\": {\"text\": $escaped_example},
      \"voice\": {\"languageCode\": \"en-US\", \"name\": \"$VOICE\"},
      \"audioConfig\": {\"audioEncoding\": \"MP3\", \"speakingRate\": $RATE}
    }" \
    "https://texttospeech.googleapis.com/v1/text:synthesize" \
    | python3 -c "import sys,json,base64; d=json.load(sys.stdin); sys.stdout.buffer.write(base64.b64decode(d['audioContent']))" \
    > "/tmp/$fname"

  # Save media to Anki + update field
  b64=$(base64 -i "/tmp/$fname")
  curl -s http://localhost:8765 -X POST \
    -H "Content-Type: application/json" \
    -d "{\"action\": \"storeMediaFile\", \"version\": 6, \"params\": {\"filename\": \"$fname\", \"data\": \"$b64\"}}" > /dev/null

  curl -s http://localhost:8765 -X POST \
    -H "Content-Type: application/json" \
    -d "{\"action\": \"updateNoteFields\", \"version\": 6, \"params\": {\"note\": {\"id\": $note_id, \"fields\": {\"Sound\": \"[sound:$fname]\"}}}}" > /dev/null

  echo "  -> TTS attached: $fname"
done < phrases.tsv

The format of the input file phrases.tsv is as follows.

Ask someone to repeat themselves when you did not catch what they said	Sorry, could you say that again?	Sorry, could you say that again? I missed the last part.	Week1
Ask someone to speak more slowly when they are talking too fast	Could you slow down a bit?	Could you slow down a bit? I want to make sure I'm following.	Week1

Summary

  • The AnkiConnect API is easy to drive with curl, enabling deck creation, note addition, and media attachment in a fully programmable way.
  • Google Cloud TTS Neural2 voices offer natural-sounding audio quality with a free tier of 1 million bytes per month, which is more than enough for personal English study.
  • Pay attention to how the Sound field is handled in card templates ({{Sound}} vs [sound:{{Sound}}]).

References

GitHub - FooSoft/anki-connect: Anki plugin to expose a remote API for creating flash cards.
Anki plugin to expose a remote API for creating flash cards. - FooSoft/anki-connect
github.com
Text-to-Speech: Lifelike AI Voices &amp; Speech Synthesis
Convert text to lifelike audio with Gemini-powered AI voices. Choose from 380+ natural-sounding voices across 75+ languages and variants.
cloud.google.com
Review pricing for Text-to-Speech
Pricing for Text-to-Speech.
cloud.google.com
Suggest an edit on GitHub
Anki Google Cloud TTS