TTS DAISY Pipeline generation - gap between sentences #813
Replies: 11 comments 2 replies
-
The gap between sentences is not fixed at 1 second, but it is fixed. The gap is determined on the level of the engines. It seems that it is generally 250ms. To control this gap, CSS isn't really appropriate, but a possibility could be to add a new setting. Another possibility could be to have the |
Beta Was this translation helpful? Give feedback.
-
Thanks for getting back to me Bert. when I measure it in an audio editing
software it says it's about 1 second. 250ms would be more natural. I am
using Azure and Ryan. I am stuck to know why it is so long. When I have
some time I will try to play with the settings.
Thanks
Paul
|
Beta Was this translation helpful? Give feedback.
-
I'll be doing some testing... |
Beta Was this translation helpful? Give feedback.
-
I am having trouble getting a small test ePub. I deleted a load of files and Sigil passes it. ACE has a few serious but not life threatening issues but DAISY Pipeline falls over! (EPUB to DAISY) [content.opf passes xml validation in NotePad++] Loading EPUB Any ideas? |
Beta Was this translation helpful? Give feedback.
-
sent direct to your gmail
|
Beta Was this translation helpful? Give feedback.
-
Thanks Bert, I have sent you directly the test ePub
Kind regards
Paul
…On Fri, 7 Feb 2025 at 14:25, Bert Frees ***@***.***> wrote:
Hmm, nothing unusual to see in the OPF file. But for some reason Pipeline
can't load it. Could you send me the whole EPUB? I need to run it and see
where it goes wrong.
—
Reply to this email directly, view it on GitHub
<#813 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AASSUUPWHZD2KJLJHNVEMUD2OS65PAVCNFSM6AAAAABWMJKBAWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMBZGQ4DQNA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
--
*Paul Wood*
Head of Technical Services
*Torch Trust*
Torch House, Torch Way,
Market Harborough, Leics. LE16 9HL, UK
Direct Line: *+44(0)1858 438269*
Mobile: *+44(0)7521 514212*
Tel: *+44(0)1858 438260*
Email: ***@***.***
Websites: www.torchtrust.org www.sightlossfriendlychurch.org.uk
<https://torchtrust.org/sight-loss-friendly-church>
Facebook: Torch-Trust
<https://www.facebook.com/pages/Torch-Trust/209414622516639>
Twitter: @torchtrust <https://twitter.com/TorchTrust>
YouTube: Torch Trust on Video
<https://www.youtube.com/channel/UCjh3k2t6SckTWOtQZNmkN8w>
[image: Walking By Faith from Torch Trust - Listen to us on your favourite
podcast platform, or here on website torchtrust.org/radio-podcasts]
<https://www.torchtrust.org/radio>
[image: Registered with the Fundraising Regulator]
<https://www.fundraisingregulator.org.uk/>
Charity No. 1095904
Privileged/Confidential Information may be contained in this message.
If you are not the intended recipient please destroy this message
and kindly notify the sender by reply email. The computer from which
this mail originates is equipped with virus screening software.
However Torch Trust cannot guarantee that the mail and its attachments
are free from virus infection.
|
Beta Was this translation helpful? Give feedback.
-
@torchtrust I used the test EPUB that you sent me, and I could convert it to DAISY without issues. I'm going to use this EPUB to do some testing. |
Beta Was this translation helpful? Give feedback.
-
OK, good! Thanks Bert.
I measure 1.38 seconds between sentences.
|
Beta Was this translation helpful? Give feedback.
-
Yes, I measure the same. I have noticed that this is only an issue with Azure voices. If I use a macOS voice, sentences are stuck together without any breaks, which is also not sounding natural at all. When I drop the break of 250 ms, the gap is only about 1.1 seconds long. This sounds already pretty natural to me (as opposed to the 1.4 which was indeed a bit long). I tried some things but removing this remaining break, which is the break that is automatically added after sentences by Azure, is not trivial. |
Beta Was this translation helpful? Give feedback.
-
I turned out to be easier than I thought. I found a way to specify the exact gap length that Azure adds. @torchtrust What would be the ideal length for you? And should it depend on the speech-rate setting? |
Beta Was this translation helpful? Give feedback.
-
Great Bert, my Audio Transcription coordinator recons:
I think the gap should vary with the speed of the reader but if not I'd
favour about 0.75s (with 140wpm).
Does that make sense?
Thanks
PAul
…On Tue, 18 Feb 2025 at 15:53, Bert Frees ***@***.***> wrote:
I turned out to be easier than I thought. I found a way to specify the
exact gap length that Azure adds. @torchtrust
<https://github.com/torchtrust> What would be the ideal length for you?
And should it depend on the speech-rate setting?
—
Reply to this email directly, view it on GitHub
<#813 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AASSUUNLRN5RWXMUZIMV7UD2QNJO3AVCNFSM6AAAAABWMJKBAWVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTEMRTHA3TSMQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
--
*Paul Wood*
Head of Technical Services
*Torch Trust*
Torch House, Torch Way,
Market Harborough, Leics. LE16 9HL, UK
Direct Line: *+44(0)1858 438269*
Mobile: *+44(0)7521 514212*
Tel: *+44(0)1858 438260*
Email: ***@***.***
Websites: www.torchtrust.org www.sightlossfriendlychurch.org.uk
<https://torchtrust.org/sight-loss-friendly-church>
Facebook: Torch-Trust
<https://www.facebook.com/pages/Torch-Trust/209414622516639>
Twitter: @torchtrust <https://twitter.com/TorchTrust>
YouTube: Torch Trust on Video
<https://www.youtube.com/channel/UCjh3k2t6SckTWOtQZNmkN8w>
[image: Walking By Faith from Torch Trust - Listen to us on your favourite
podcast platform, or here on website torchtrust.org/radio-podcasts]
<https://www.torchtrust.org/radio>
[image: Registered with the Fundraising Regulator]
<https://www.fundraisingregulator.org.uk/>
Charity No. 1095904
Privileged/Confidential Information may be contained in this message.
If you are not the intended recipient please destroy this message
and kindly notify the sender by reply email. The computer from which
this mail originates is equipped with virus screening software.
However Torch Trust cannot guarantee that the mail and its attachments
are free from virus infection.
|
Beta Was this translation helpful? Give feedback.
-
Hi,
I am back on the TTs engine in DAISY pipeline again. I can control the gap between the paragraphs and headings using css:
h1 { pause-before: 1500ms; volume: x-loud }
h2 { pause-before: 1000ms; volume: loud }
book {speech-rate: -12% }
p { pause-after: 1100ms }
p.quotation { pitch: high }
but can we control the gap between sentences? is it fixed at 1 second? I am finding it a little long and I would like to reduce it so the speech flows more naturally.
Thanks
Paul
Beta Was this translation helpful? Give feedback.
All reactions