Integration with Speechmatics

[rkpujari]

The editor looks great. Thanks for this. I am looking for Speechmatics integration in this. Any plans to add that one? if not, some documentation on how to integrate it into this editor will be helpful. Thanks in advance.

[pietrop]

Thanks @rkpujari , The idea is to make it easier to create adapters and converters to import and export data into the TranscriptEdior component.

We are planning to write some more documentation and guides on how to do that. Should have something in the next couple of weeks.

For now the adapter folder is in /src/lib/TranscriptEditor/TimedTextEditor/adapters and you can see how the BBC Kaldi or the autoEdit one are done, to see what it would take to make a Speechmatics one.

But, yes, will update this issue with more info soon.

[pietrop]

@rkpujari we have a first draft of a PR #51 with some instructions on how to create adapters for various STT services.

We are still tweaking it and reviewing it for clarity etc.. but feedback welcome if this is something you are currently looking at.

[rkpujari]

Hi, Thanks for the update. The documentation will be very useful for me. I added the adapter for Speechmatics already though I need to make some enhancements to the editor and working on that. I need to show speaker names (option to turn on/off), readjust timestamp, highlights words having low "confidence", export the transcript to Word format, font size change, add spell check (and suggestions for alternate words) for newly added text in the transcript etc. Thanks for your help. regards, Rama

…

On Fri, 14 Dec 2018 at 20:46, Pietro ***@***.***> wrote: @rkpujari <https://github.com/rkpujari> we have a first draft of a PR #51 <#51> with some instructions on how to create adapters for various STT services. We are still tweaking it and reviewing it for clarity etc.. but feedback welcome if this is something you are currently looking at. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#38 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ArYbVTBXaMOy7mo1IW8R0aGW-4Qj0kbwks5u48DGgaJpZM4Y7qzP> .

[pietrop]

If you want to do a Pull Request with the Speechmatics adapter we can add it to the main component.

We are also working on

• having the possibility to turn speaker names and time stamps on or off.
• readjust time stamps ( word timings after text has been edited)
• hilight words having low confidence ( below 6)
• export transcript ( for now as plain text)
  ....

[rkpujari]

Good to know that you are working on those features (interested to know if that can be expected very soon or may take time). regarding the pull request for Speechmatics adapter, I am not an expert React developer. I copied the BBC Kaldi adapter and made minor changes to it to get it working. attached the index.js file (converted to text file to stop mail scanning). Thanks, Rama

On Tue, 18 Dec 2018 at 15:01, Pietro ***@***.***> wrote: If you want to do a Pull Request with the Speechmatics adapter we can add it to the main component. We are also working - on having the possibility to turn speaker names and time stamps on or off. - readjust time stamps ( word timings after text has been edited) - hilight words having low confidence ( below 6) - export transcript ( for now as plain text) .... — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#38 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ArYbVewb6JXaoFBm1z_XkiMEJPrmswoXks5u6LYNgaJpZM4Y7qzP> .

/** * Convert BBC Kaldi json ``` { "action": "audio-transcribe", "retval": { "status": true, "wonid": "octo:2692ea33-d595-41d8-bfd5-aa7f2d2f89ee", "punct": "There is a day. About ten years ago when ...", "words": [ { "start": 13.02, "confidence": 0.68, "end": 13.17, "word": "there", "punct": "There", "index": 0 }, { "start": 13.17, "confidence": 0.61, "end": 13.38, "word": "is", "punct": "is", "index": 1 }, ``` * * into * ``` const blocks = [ { text: 'Hello', type: 'paragraph', data: { speaker: 'Foo', }, entityRanges: [], }, { text: 'World', type: 'paragraph', data: { speaker: 'Bar', }, entityRanges: [], }, ]; ``` * */ import generateEntitiesRanges from '../generate-entities-ranges/index.js'; /** * groups words list from kaldi transcript based on punctuation. * @todo To be more accurate, should introduce an honorifics library to do the splitting of the words. * @param {array} words - array of words opbjects from kaldi transcript */ const groupWordsInParagraphs = (words) => { const results = []; let paragraph = { words: [], text: [] }; words.forEach((word) => { var endTime = parseInt(word.time) + parseInt(word.duration); const tmpWord = { text: word.name, start: word.time, end: endTime, }; // if word contains punctuation if (/[.?!]/.test(word.name)) { paragraph.words.push(tmpWord); paragraph.text.push(word.name); results.push(paragraph); // reset paragraph paragraph = { words: [], text: [] }; } else { paragraph.words.push(tmpWord); paragraph.text.push(word.name); } }); return results; }; const SpeechmaticsToDraft = (SpeechmaticsJson) => { const results = []; let tmpWords; console.log("entered speechmatics adaptor"); tmpWords = SpeechmaticsJson.words; const wordsByParagraphs = groupWordsInParagraphs(tmpWords); wordsByParagraphs.forEach((paragraph) => { const draftJsContentBlockParagraph = { text: paragraph.text.join(' '), type: 'paragraph', data: { speaker: 'TBC', }, // the entities as ranges are each word in the space-joined text, // so it needs to be compute for each the offset from the beginning of the paragraph and the length entityRanges: generateEntitiesRanges(paragraph.words, 'text'), // wordAttributeName }; // console.log(JSON.stringify(draftJsContentBlockParagraph,null,2)) results.push(draftJsContentBlockParagraph); }); return results; }; export default SpeechmaticsToDraft;

[pietrop]

ok, that's great, thanks for sharing this, I think with your code example, and the docs from speechmatics it should be possible to add it as a separate adapter, following the guide in the PR #51 - Guide: How to Create an Adapter - Draft

[pietrop]

In terms of time estimate @rkpujari as an update we just got the hide/show time-codes and speaker label functionality working in PR #56 and we should be able to add that to master soon.

re-aligning text I reckon end of January / February, to give a conservative estimate🤞

Altho this might not be needed for most use cases, where users are just correcting and exporting text (?)

[rkpujari]

Thanks a lot for your help. Looking forward for the latest version. regarding realigning the text, sometimes particular section or words of transcript could be inaccurate due to poor audio quality segment (or due to an issue with the speech conversion tool) and in that case if user manually replaces the transcript with new text, then the user should be able to associate the timestamp to the newly added text in the transcript (so that it gets highlighted when audio is played and vice versa). I don't have such audio and sample transcript for such audio and so I am not sure how transcript will look like in that case and how editor works in that case. I understand your comment that the use case may not be needed since user will be correcting and exporting it. need to speak with my product owner to get more details. I have seen the new UI for the editor. looks very good and there is more space for the transcript now. Thanks once again. regards, Rama

…

On Thu, 20 Dec 2018 at 19:54, Pietro ***@***.***> wrote: In terms of time estimate @rkpujari <https://github.com/rkpujari> as an update we just got the hide/show time-codes and speaker label functionality working in PR #56 <#56> and we should be able to add that to master soon. re-aligning text I reckon end of January / February, to give a conservative estimate🤞 *Altho this might not be needed for most use cases, where users are just correcting and exporting text (?)* — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#38 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ArYbVcTFCsMegKqTMa3E5C39MfCNI3i8ks5u652TgaJpZM4Y7qzP> .

[pietrop]

Thanks @rkpujari ,

Yes, that is correct, if the user changes incorrect words, (especially if they completely delete a word or a paragraph) then the time-codes will no longer be associated with the text (and they can no longer click on that text to jump to the corresponding point in the media).

And just to clarify, we are exploring ways to realign text, I've written some notes on some progress we have made on that front here #30 we should be able to try this out within the editor in the new year to be able to do some more comprehensive test, and see if it's a valid solution.

[murezzda]

I'm also very interested in this project and in the speechmatics adapter for this tool. What are your current plans for officially adding the support? I've already wrote a speechmatics adapter for getting to know the code. I would happily share and provide a pull request if you are interested.

[pietrop]

Thanks for reaching out @murezzda, yes a PR for the Speechmatics adapter would be great, thanks!

You can see this guide for how to add a new adapter as well as the contributing section section).

And feel free to reach out with any questions you might have.

[pietrop]

Addressed in PR #94

[pietrop]

from version @bbc/[email protected] there is a speechmatics adapter as part of the component, thanks to @murezzda