We have added Norwegian (Bokmal), Swedish and Dutch to the Transcription Panel in Premiere Pro. Since I don't speak any of these languages I would love to know how well they are working for you. Please try some of your footage and let us know: What kind of content are you transcribing? (zoom call, promotional video, YouTube DIY tutorial, Netflix documentary...) Does it do better with certain dialects? If there are regional dialects that it doesn't handle as well, be specific about the names of the region, accent or dialect. How accurately does it transcribe numbers? (1984, February 1st) How accurately does it handle pop culture (movies, current events, famous people)? Will you use it in its current state?

Community Manager

Question

Speech to Text Adds Norwegian, Swedish and Dutch

Forum|Forum|3 years ago
May 19, 2022
26 replies
18688 views

We have added Norwegian (Bokmal), Swedish and Dutch to the Transcription Panel in Premiere Pro. Since I don't speak any of these languages I would love to know how well they are working for you. Please try some of your footage and let us know:

What kind of content are you transcribing? (zoom call, promotional video, YouTube DIY tutorial, Netflix documentary...)
Does it do better with certain dialects? If there are regional dialects that it doesn't handle as well, be specific about the names of the region, accent or dialect.
How accurately does it transcribe numbers? (1984, February 1st)
How accurately does it handle pop culture (movies, current events, famous people)?
Will you use it in its current state?

Feedback

This topic has been closed for replies.

S

Stianlud

Participant

I tested this on a norwegian gaming clip and I must say there are a lot of fails. Like "der e han der e han" or "there he is there he is" got texted to "sa til dere jenter" or "told you ladies". There was also many random Norwegian places just random placed in the transcript. I and the other guy in the clip are from Bergen, Norway and talk a dialect called "Bergensk" and it definitely need more training on that. I can help with some data or something if you should want 🙂

R Neil Haugen

Legend

Best wishes! It's a pretty language to hear, but my oh my, does it have it's intricacies ...

Everyone's mileage always varies ...

S

Stianlud

Participant

Absolutely, I think the model must be trained on a lot of dialect data if it is to be able to master Norwegian in a good way. Especially "Bergensk"

J

Jarle Leirpoll

Legend

I've now tested Norwegian transcription on a lot of different files. I'm afraid it's still not very good (23.4 beta, build 37).

It often has problems with numbers.
It splits words like it was English, while the Norwegian language tends to join words.
(I think the English terms are Compound words. English uses open compounds a lot, while Norwegian uses closed compounds.)
It struggles a lot with Norwegian dialects - and we have many dialects, used by a huge part of the population.
It often wants to put a period, followed by a capital letter, in places where they don't belong.

It's not unusable, because it gets quite a substantial part of the transcript right - but it's nowhere near the quality I get with English transcription.

R Neil Haugen

Legend

Somehow, I'm not surprised at your experience. English based operations, going into Norsk ... that's difficult, and not just because of the (very literally) 654 recognized dialects at this time in Norway. Which is down from the 750+ of back around 1990.

Having some capabilities in American (US English, which ain't what them Islanders on the other side of The Pond mutter ... ), Deutsch (German (sigh) ... for most ... Americans ... ) and the 'home family' language of Norsk, I tell people the following:

Say you have a typed paragraph that in American English is seven lines. In the same font, in Deutsch, it is probably around ten. And in Norsk, it is probably no more than five. As in Deutsch things are specified in great literal detail, but i Norsk, so many things are "understood ... " from the context of the phrase, without the written specifics.

Now add in the variety of spoken dialects in Norsk. And that in English, nouns aren't normally "gendered". "The" and "a" are used for all nouns the same way. The table, the car, the ... anything. A table, a car ... same thing.

But in Norway, in most of the country there are two different noun genders with the accompanying declensions/tenses/whatever. As in "et bord" (a table) but "en bil" (a car).

And the "the" part is suffixed in other uses such as plurals ... "bordet" for plural tables (general), "bilen" for a general group of cars. But in some significant parts of the country there are THREE noun genders ...

And then, Norsk is in some ways a simple language, but uses so many "understood ... " phrasings, with fewer actual words than would be used typically in English. That alone makes it a complex thing to do written translations.

Now have an English based group designing a Norsk transcription service ... oh, I see room for problems all over ...

Neil

Everyone's mileage always varies ...

J

Jarle Leirpoll

Legend

I know. We're screwed 🤷‍ 🤣

S

sondreterjesen

Participant

Are these languages still available? I'm making a lot of cooking videos these days and I spend alot of time trascribing norwegian. Would be great to have a go at this.

Stan Jones

Community Expert

I don't recall when it was added, but Norwegian is in the Release 23.2 version (not Beta).

Stan

S

sondreterjesen

Participant

Thanks 🙂 Found it after an update

N

NwProPr

Known Participant

Hi,

I've tried the Dutch language pack several times in the last few weeks (on the latest betas), but to be honest, it's not the greatest. There are multiple factors which can break the entire thing (dialect, a lot of specific words, pauses, thinking out loud etc).

I was wondering how Adobe is improving these language models. Are there ways where we, Dutchies, can help?

Having transcribed over 50 different interviews, it only turned out useful for half of the sequences. The interviews where recorded with lavalier mics, so the audio quality is just fine and most speakers were very clear in terms of pronunciation audio quality.

One thing that would be useful is to add a delete option. So you can delete transcriptions from a sequence (instead of having to create a new sequence and copy all clips). I've noticed that the more sequences are stored in a single project with large amounts of text, the slower the sequences get.

Neckebba

Inspiring

I have a total question : the Adobe captions for dutch. How conitnious will they be updated ? Like i know/feel this is work in progress. And is still in full developement and not nearly perfect. Other online services are doing a continoius updated on this. How is Adobe have a look at this? Like we can add language pack. but if we stay on that version , that language pack stays same . Are there add ons? expansions? updates on this ?

Stan Jones

Community Expert

A good question.

At some point we were told where the language pack resides. But not how/how often it may be changed. But the language pack files are installed per version (e.g. 22.0, 23.0, Beta 23.0). And the "pack" is a bunch of files. I assume they will update no more often than with program updates. And with Beta, they would update if there was a change for that build.

Stan

K

Kees Veling

Inspiring

I tried it on an edited piece, interviews with educators from all over the Netherlands speaking regular Dutch with light accents. Closely miked, no echo or noise at all.

It works really well, i have been using microsoft word's online captioning to a word document, this works much better.

I found the following in this short test:

-The captioning should leave out "uh"and "uhm"as there rarely add meaning, manually adding it every so often is much easier than deleting hundreds of useless uhms.

-The " 't " sound is being captioned as "H " it should be captioned as "het " or " 't "

-I t would be wonderful if it could follow the Dutch guidelines for subtitles when breaking up sentences and use the ... dots.

-It capitalizes new sentences without adding a dot to end the sentence before it.

-In breaking up lines it should try and allways have more than one word on each line.

S

Sigurd28177789locn

Participant

Hi! Where can I find this package? I doesn't show up in the transcription panel in my program.

S

Sven5FFB

Participant

Hi Teresa, i looked for the Dutch transcript option but can't seem to find it. I have Premiere Pro 2023 installed on iMac Intel version.

W

Wessel26409343rd9u

Participant

I'm gaving the same issue. I can't find the dutch transcript language pack anywhere. Maybe they removed it for some reason? Hope this gets added soon!

Ann Bens

Community Expert

S

Samarbeisråde28116854iqtt

Participant

I for some reason cannot create captions in the Beta-version. work fine in the regular version. I also cannot change language after I have selected norwegian. The meny that usually pops up and lets you select language does not show up anymore. Tried different projects, files, and of course turning on and off. Anyone run into this, or know how to fix it? 🙂

The transcript itself is pretty good though, not so good with mentioning years (2022 turns into 20 to, and "år"=year turns into "ord), but works pretty great.

Stan Jones

Community Expert

Samar,

To get the language options, in the generate transcript window, you must click on the little triangle next to "Preferences (language, labeling, audio).

I have not seen the "Create Captions" button greyed out.

What specific version of the Beta are you using? Are you clicking on the CC icon in the Transcript tab?

Stan