Hyakunin Isshu reader: make karuta reading audio for practice
Karuta needs a reader: someone reads a poem from the Hyakunin Isshu and players race to take the card with its second half. Practising alone or in a small club, there is often nobody free to read, and recordings do not always come in the order or pace you want.
This guide explains the reading timing used in competitive karuta, how to write the 100 poems so a voice reads them correctly, and how to voice all of them in Japanese with the free SuperClonica engine for a shuffled practice track.
How a karuta reading sounds
According to the All-Japan Karuta Association’s reader training, the basic pattern is called 5-3-1-6: the lower verse of the previous card in about 5 seconds, a held final sound of 3 seconds, a 1 second pause, then the upper verse of the next card in about 6 seconds. A match starts with the opening poem “Naniwazu ni”, which is not one of the 100 cards. A synthetic voice reads the words clearly, but it does not sing or hold the last sound the way a trained reader does.
Write the poems as they are read
Classical spelling is not read as written: けふ is read きょう, てふ is read ちょう, and 庵 in the first poem is いお. Write each verse in hiragana the way it is pronounced, with 、 between the phrases, for example あきのたの、かりほのいおの、とまをあらみ for the upper verse of poem 1 and わがころもでは、つゆにぬれつつ for its lower verse. The poems are over 700 years old and free to use.
Voice all 100 cards in Japanese
Install the free engine with python -m pip install superclonica-cpu and voice the upper and lower verse of each poem into separate files, for example superclonica --text "あきのたの、かりほのいおの、とまをあらみ" --voice voice.json --lang ja --output 001_kami.wav. Use the same voice and seed for all 200 files. If a verse comes out much shorter than the 5 or 6 seconds a reader takes, slow it down with --speed.
Build a shuffled practice track
Shuffle the 100 numbers, then join the files in order: the lower verse of the previous card, about 4 seconds of silence for the held sound and the pause, and the upper verse of the next card. Start the track with the opening poem. Any audio editor or a short script can do this, and you get a new order every time you shuffle.
Choose the voice
Try voices free on their pages with one verse, 30 to 100 characters, no account needed. Catalog voices were recorded in English and Russian and keep some accent in Japanese, so listen before you choose. For a native reading, a club member who agrees can record 5 to 10 seconds of clean Japanese and train a voice in the studio. For tournaments, follow the rules of your organiser: official matches use human readers.