Opinions, tips, and news orbiting Microsoft
Automate Grading & Personalize Feedback. OK autogrades programming assignments, facilitates submission, composition feedback, and analytics for your class. View a Demo of OkPy The ok.py server collects submissions and displays analysis of student progress based on logging sent from client scripts the ok.py software was developed for CS 61A at UC Berkeley and is now… Continue reading Deploy OkPy to Azure PaaS
First published on MSDN on Jan 25, 2019
Authored by Peter Laker
Feed audio from friends, family or vids of your favourite celebrity.
Train a customised voice that sounds just like them.
Then make them say whatever you want!
[audio wav="https://msdnshared.blob.core.windows.net/media/2019/01/Martin-Fake-Voice-V2.wav"][/audio]
OK so this still sounds a little bit robotic. But the training data was less than 8 minutes of poor quality audio, with non-matching text.
With not much more effort, you can buff this up to sound very close to the real person!
To hear what Martin sounds like (and learn a thing or two)
watch this excellent series of clips about Azure DevOps
.
This is a rapidly maturing technology, which has many real-world benefits. You can generate custom voices for chat bots, to use in games, for interacting with home automation, including Cortana, Alexa or Siri skills. It offers very believable video narration, and more realistic voice for call handling, to name just a few popular uses. The transcription service was where I came in. It is driven by the latest AI from Azure Cognitive Services. It provides almost real-time transcription and corrects itself as it goes.
As an IT pro reading the docs, training a “Custom Voice” with Microsoft Speech Service
sounds
like a complicated and cumbersome task.
I also initially thought it sounded like too big of a job, to prepare all the training data for a decent sounding voice.
But it wasn’t, and
I hope to dispel that myth for you as well
!
The services are very forgiving and will produce recognisable results with under a minute’s worth of training data!
The code I’ve
uploaded to GitHub
makes it even simpler.
Instead of spending hours reciting carefully scripted phrases for training data, why not flip it around?
Take existing audio and first transcribe that to text.
Then feed that the original audio and the auto-transcription BACK into the service!
For even better results, spend as little as half an hour, tidying the transcription mistakes and feed that curated/cleaned version back in.
You’ll see how easy that is, with the demo tool I’ve posted
posted to GitHub
.
It also helps you splice the audio up, into small chunks to feed into the service.
This visual tool will do everything required to prep the training data and you can learn and adapt it to your needs.
Read more and watch demos of this project
HERE
This project (which you can just download and run) will show you how to authenticate against the Speech Service, how to upload audio for transcription and also how to request and download the your final custom generated audio files.
You can
download the code for this project from GitHub
and have your own custom voice ready, in under an hour!
The initial results may initially sound a little “robotic”, of course. But it’s shocking to hear a familiar voice coming back at you. The final quality of your voice is actually only limited to the amount of time you then have. To tidy up the samples, perfect the acoustic library and work on the phonetic scripts. But you can still get very impressive results, with very little work.
Simply for better transcription results. I initially just wanted to provide viewers a transcript of a video presentation that
Martin Woodward
(
Microsoft Principal Group Program Manager for Azure DevOps
) gave, to my
Azure Developer Meetup
group last November. You can watch the full set of eight clips from that great session here.
The basic recognition code was
copy n paste from the docs
and I had transcription results within just FIVE minutes! However, the sound quality and his slight accent meant the transcription had many errors. they would have taken me some time to manually correct.
So I decided instead to feed the “best bits” of the audio BACK into the service, generate an acoustic model and hopefully get a much improved transcription out.
And it worked!
As a side-effect, I also generated a custom voice, which was trained to sound very similar to Martin! I was so impressed at how easy it was.
So over the festive break, I pulled together some demo code to show how it is implemented.
This is now available in
this C# GitHub repo
, to demonstrate how to implement this.
You can download that now and try yourself.
It’s easy!
Within just 30-60 minutes, you can get impressive results!
I’m filling out
a more detailed article on using the service HERE on TechNet Wiki.
Plus, I’ll be documenting the project features in the GitHub project wiki – if you just want to use the tool and not know the code.
Pete Laker,
Azure MVP
and
Microsoft IT Implementer
This is part of a series of Azure developer blog posts, in my role as
Microsoft IT Implementer
.
I’m honoured to have been asked by Microsoft to demystify and demonstrate some of the latest technologies, which excite me – and I hope you too!
First published on MSDN on Jan 17, 2019
Authored by Sjoukje Zaal
Welcome to another Azure Content Spotlight! These articles are used to highlight items in Azure that could be more visible to the… Continue reading Azure Content Spotlight – Azure Data Architecture Guide
First published on MSDN on Jan 14, 2019
Authored by Andreas Helland
Talking about containers is getting old by now, but in the light of my last article on deploying to Docker Hub via Azure DevOps I still thought it… Continue reading Playing with Containers in Azure App Service
First published on MSDN on Jan 07, 2019
Authored by Andreas Helland
I frequently write code that is meant for public consumption. It might not be perfect code, but I make an effort to make it available for others t… Continue reading Setting up a Docker Containers Infrastructure with Azure DevOps
First published on MSDN on Dec 29, 2018
Authored by Stephane Eyskens
Azure Security Kit aka AzSK is a framework that is used internally by Microsoft to control & govern their Azure Subscriptions. While so… Continue reading Enforcing security controls right from CI/CD pipeline with AzSK – Deep Dive
Applies to: Project Service Automation for Dynamics 365 We’re pleased to announce the latest update for the Project Service Automation application for Dynamics 365. This release includes bug fixes. This release is compatible with Dynamics 365 for Customer Engagement apps version 9.x. To update to this release, visit the Admin Center for Dynamics 365 online,… Continue reading Release Notes for Project Service Automation Update Release 12, V2.4.9.70
Today, the Microsoft DevOps blog has began moving to a new platform with a modern, clean design and powerful features that will make it easy for you to discover and share great content. You’ll see the new blog live in the next few hours! The DevOps blog, along with a select few others (Visual Studio… Continue reading Microsoft DevOps blog has moved!
