The Will to Resist

Jerry's Marginalia — Apparently I Have a Transcription Department Now

Technology occasionally solves a problem you did not realize you were going to have.

This is dangerous.

Because once Jerry discovers that a thing is possible, Jerry immediately begins looking for increasingly stupid applications of that thing.

Today's culprit is Any2Text.

I originally needed to turn some very long audio recordings into text.

Not exciting.

Not glamorous.

Certainly not the beginning of an international localization empire.

I uploaded the audio.

It transcribed it.

This was useful.

Unfortunately, Jerry then noticed that the service supports multiple languages.

Including Japanese.

This produced a thought.

There are old anime, OVAs, interviews, games, and assorted pieces of Japanese media floating around the Internet that have never received an English translation.

For decades, the answer to this problem was straightforward:

Learn Japanese.

This remains an excellent solution.

It is also somewhat time-consuming.

But now we live in the future, where software can potentially perform the ugly first step: listen to hours of Japanese audio and produce a transcript.

A transcript is not a finished translation.

A machine translation is not a finished localization.

And whatever an AI produces should never automatically be trusted simply because the computer said it confidently.

But give Jerry a Japanese transcript, a rough translation, enough context, and an editing pass?

Oh no.

Germlin Subs has entered pre-production.

Translation: Germlin

Editing: Germlin

Timing: Nyx

QC: Cactus

Special Thanks: A2T

None of these people should be trusted.

For what I actually needed, though, Any2Text has been surprisingly useful. I have already thrown several recordings at it, including one lasting more than five hours, and being able to search and read a transcript instead of repeatedly scrubbing through an MP3 is enormously convenient.

It isn't magic.

Audio quality matters.

Speech recognition can misunderstand people.

Names can become casualties.

And any transcript important enough to rely upon should still be checked against the original recording.

But as a tool for turning an enormous wall of audio into something I can actually navigate?

Yeah.

I can see myself using this.

The Premium plan currently advertises 3,000 minutes per month for $19.99, which is fifty hours of transcription.

Jerry should not be given fifty hours of transcription.

Jerry has ideas.

Terrible ideas.

Potentially subtitled ideas.


Disclosure: Any2Text offers 100 transcription minutes for publishing an opinion about the service and sending them a link. I wanted the free minutes, so here we are. They did not tell me what to write or require this to be positive.


Jerry “The Ankle-Biter” Silverhand


Marginal note: QC assures me the subtitles are accurate. Unfortunately, QC is apparently named Cactus. 🌵