8
My new thing: automated transcription and dialogue-only audio (sh.itjust.works)
submitted 5 days ago by schipelblorp@sh.itjust.works to c/languagelearning@sopuli.xyz
4 comments fedilink hide all child comments

A few weeks ago, I asked y'all where you got your subtitles. Someone recommended Bazarr, but I became intrigued by the idea of automated transcription. Subtitles are often summaries of the dialogue, and often use different vocabulary and construction, so I thought if I could get an automated transcription of any show, that would be the bee's knees.

So my new pipeline:

Show --> automatic transcription into SRT file --> manual correction & touch up --> compressed audio file of dialogue only

Automated transcription is... OK. There's plenty of stuff that it misses, so it's important to be listening the first time through for hallucinations and mistakes; otherwise you waste your time looking up irrelevant grammar and sabotage yourself listening for things that aren´t there. I don´t know if I'd recommend it to someone just starting out, but if you're phonetically aware to at least realize that the text does not match what you're hearing, I think automated transcription is ok.

And once you have a transcription keyed to the actual spoken dialogue, you can create a dialogue-only audio file, which means that instead of listening to a 100 minute movie for 20 minutes of dialogue, you can listen to 20 minutes of dialogue 5 times.

Anyone do similar? Tips? Tricks?

Details: faster-whisper with VOD enabled.

top 4 comments
sorted by: hot top new old
[-] emb@lemmy.world 3 points 5 days ago* (last edited 5 days ago)

Subtitles are often summaries of the dialogue, and often use different vocabulary and construction

That drives me crazy! Especially if I'm watching something translated, pretty often the subtitle and VO track will be completely different.

Usually I get around this by watching something originally made in the language I want to hear. Then the subtitles tend to be a closer match.

Haven't tried auto-transcription. It sounds promising, but also like a lot of work. With that manual cleanup step in the middle, seems like you'd already need a good understanding. But maybe that work would be pretty educational itself, good practice.

[-] schipelblorp@sh.itjust.works 3 points 5 days ago* (last edited 5 days ago)

Yes, I started with the Simpsons and I swear the Spanish subtitle and dub teams had some kind of bet they could do an entire episode without sharing a single word between them.

A pretty minimal way to use even inaccurate transcription is to mine it for vocabulary. When you do hear a word you know, you have a better chance to recognize it. If the word is transcribed wrong (and you know what it sounds like), you're not going to hear it in the audio, and there's no harm done but perhaps you learned a word you didn´t need right away.

What you definitely don't want to do is use the transcription as a pronunciation guide before you have any idea of what anything is supposed to sound like.

But if you're input heavy like me and would kind of listen to something you understand 30% of that is mildly interesting instead of something you understand 90% of that is interminably boring (like most slow-version audio), it's worth considering.

It is a lot of work to set up, but it's frontloaded. Once you have everything installed and know how it works, you can spend more time on language learning and have total freedom in your sources.

Edit: I'll also admit I used an LLM to help figure some things out, including coding, but I have done a little programming and use the CLI linux so I wasn´t completely blind. Use at your own risk.

[-] csolisr@hub.azkware.net 2 points 5 days ago

I swear the Spanish subtitle and dub teams had some kind of bet they could do an entire episode without sharing a single word between them.

Yeah, there are two good reasons for that: lip sync, and adapting references to the local audience. The subtitles generally don't have to bother doing either of those, and furthermore they must limit their line size to make the subtitles readable at a normal speed. The result is that the sub and the dub diverge greatly - back when I still watched media in Spanish, I used to put subs and dubs on simultaneously to catch as much information from the original as possible

[-] schipelblorp@sh.itjust.works 3 points 4 days ago

I am aware. It was short phrases that could easily have fit into subtitles unchanged that largely drove this suspiscion that there was some purpose, some consciousness, behind the divergence.

this post was submitted on 06 Sep 2026
8 points (90.0% liked)

Language Learning

1012 readers
7 users here now

A community all about learning languages!

Ask / talk about a specific language or language learning in general.

Sopuli's instance rules apply

  1. Remember the human! (no harassment, threats, etc.)
  2. No racism or other discrimination
  3. No Nazis, QAnon or similar whackos and no endorsement of them
  4. No porn
  5. No ads or spam
  6. No content against Finnish law

Other active Lemmy language communities:

Other communities outside Lemmy:

Community banner & icon credits:

Icon: The book cover of Babel (2022 novel by R. F. Kuang)

Banner: Epic of Gilgamesh tablet (© The Trustees of the British Museum)

founded 3 years ago
MODERATORS
Lazycog@sopuli.xyz
emb@lemmy.world