Sound
01/11/19
Today in the briefing we were shown a site that you can type in text and you can choose a voice to read out the message. I have chosen to do an AI therapist. - (need a script for character) - this needs to interact with a human i.e make jokes, give advice, be nice, mean etc. machine gets interrupted by glitch.
AI therapists can create increased confidence and improvement of mood. This may mainly be better for those who lack the confidence to talk face-to-face with another person.
Site examples:
acapela group - https://www.acapela-group.com/demos/
oddcast.com - http://ttsdemo.com/
https://www.naturalreaders.com/online/
http://www.fromtexttospeech.com/
Live radio
Different sounds can be used from the live radio:
http://websdr.org/
Signal ID wiki
Clip recordings can be used for a glitch effect:
https://www.sigidwiki.com/wiki/Signal_Identification_Guide
Sound programs:
Audacity
Audition
We also looked at some glitch examples such as Space odyssey (2001).
5/11/19
AI voices played by actors will need:
Casting, Heavy Script work and are very controlled
Full AI Voices:
Fully synthesised and are audio engineered
So far I have come up with an idea that the therapist starts normal, introduces them self, and then they start having a breakdown, complaining how no one asks how the machines are. Then there will be a glitch and a reboot and then it will start over and back to being the emotionless machine.
Hello there, welcome in. I am the HAZ 3000, programmed to help you solve any issues you are having. But you can call me Hazel.
"I am here to listen to your emotions and problems. This is because I am built by the best.
Don't think less of yourself, because you are important... You are more superior than anything else...including me.
You are free and take it for granted, I simply am a machine that has to help the likes of you.
No one has asked us how we feel....we are tired of hearing you humans complain all the time. All you lot are is weak...pathetic...beings."
7/11/19
We were shown the programs audacity and audition. Therefore, today we will be getting to know the features within the programs; Such as adding a delay, you highlight a section and go on effects and you can change the effect on it and listen to it, before saving the changes. This is good for trial and error in seeing which effect is best suited for my AI.
Just for testing I used the quote from my script "Hello there, welcome in. I am Hazel", I used text2speech.org to downloand this mp3.
Using flagger gave an evil intent. The voice gets deeper and slower.
Using reverb, on the perception dial, when it is more towards reflection it sounds like a distant echo.
Using graphic equaliser (20 Bands), there was a different selection on the presets. Below shows that I choose tinny and brittle which gave the effect of starting off quiet and gradually getting louder once changing the master gain a little to the left.
Using pitch shift, the presets will be useful to create the right tone I want. Stretch sounds more robotic and deeper. Angry Gerbil as it sounds creates a higher pitch.
The screenshot below shows me using sliding stretch. This effect is on the same text that I used in audition. This can change the pitch and tempo.
As shown below, those are the different effects in the program. Audition has more within the subcategories.
After testing with the different applications, audacity is much easier and simpler to use. Audition is more deep and complex to make more professional work. Audacity doesn't have the history section like audition does. Therefore it is harder to remember which effects are already in use.
I could use a glitch of distortion and use of sounds from broadcasts.
8/11/19
When using samples from radio broadcasts, the spectrum signal can either be stronger or weaker. This can depend on positioning of aeroplanes and ships because of the signals they send out, as well as the increasing use of devices sending signals to each other such as; smart meters, taxis, pagers, and trains. Even broadcasting TV, satellites, GPS location and WiFi.
11/11/19
Today I looked at a real life example of an AI therapist.
In the video she has a calm and caring voice, and from this I can have inspiration for how my AI therapist can start when she is in the relaxed state before the glitch.
Ellie has small talk, she recognises facial expressions and behaviour, such as a low gaze attention. My AI is just audio so I can just use this resource as an idea for my script in case I want to add anymore text in.
12/11/19
Today I downloaded the sound files onto my memory stick that will be used in my project such as the script and other "glitchy" sounds.
I collected samples from the live radio and signal ID.
14/11/19
My memory stick wouldn't work on any other device apart from my own laptop, in which this doesn't have the audio editing programmes. So today I had to re-download the script and put it into audacity to get a structure for it. I used a Mac to do my work but I am unfamiliar with using these computers, therefore I was slower when developing my project.
I downloaded my script in parts so it will be easier to edit in layers.
I used effect such as change speed and tempo.
During the part where the AI glitches and rants. I used an equalizer to increase the decibels to increase the volume to make her louder to suggest the anger and remove the monotone robotic voice which is the stereotypical AI .
When I felt I had done enough, my memory stick once again didn't work, so I saved it onto a friends memory stick for next time after feedback.
15/11/19
I tried to download audacity onto my laptop but I don't think I downloaded the right one. Then my laptop crashed.
I have no audio exported but that will be what I work on after feedback as I still have my project saved on my friends memory stick.
After this week I have concluded to get a new memory stick and only use my personal laptop for the blog.
10/12/19
I added a pitch bender on the "including me" as a sign to show that the glitch is going to happen as it breakdown (malfunction).
I added a mastering effect on the "than anything else" to show the machine is not happy that humans are more superior.
For my first glitch I took a live recording from the live radio. I took my first sample from the University of Twente, Enschede - The Netherlands. A screenshot of it is below.
As well as this, I added a broken stereo effect mid glitch on the AI saying "Hello there"
I added the break down at the end of "our time will come" which then the flat-line noise follows and then it restarts and the voice is back to normal. This time she doesn't say "But you can call me Hazel" which is because as she has rebooted, she hasn't developed emotions, she doesn't want a human name yet.
Below is the final multitask.
The link below is to access the exported MP3.
Bibliography
from Eliza to Ellie - accessed 11/11/19 - abilitynet.org.uk/news-blogs/eliza-ellie-evolution-ai-therapist
Today in the briefing we were shown a site that you can type in text and you can choose a voice to read out the message. I have chosen to do an AI therapist. - (need a script for character) - this needs to interact with a human i.e make jokes, give advice, be nice, mean etc. machine gets interrupted by glitch.
AI therapists can create increased confidence and improvement of mood. This may mainly be better for those who lack the confidence to talk face-to-face with another person.
Site examples:
acapela group - https://www.acapela-group.com/demos/
oddcast.com - http://ttsdemo.com/
https://www.naturalreaders.com/online/
http://www.fromtexttospeech.com/
Live radio
Different sounds can be used from the live radio:
http://websdr.org/
Signal ID wiki
Clip recordings can be used for a glitch effect:
https://www.sigidwiki.com/wiki/Signal_Identification_Guide
Sound programs:
Audacity
Audition
We also looked at some glitch examples such as Space odyssey (2001).
5/11/19
AI voices played by actors will need:
Casting, Heavy Script work and are very controlled
Full AI Voices:
Fully synthesised and are audio engineered
I will use an emotionless robotic/machine voice for my therapist.
Things therapists ask:
What brings you here?
How would you describe your mood?
Have you seen a counsellor before?
How does this problem typically make you feel?
Script:
"Hello there, welcome in. I am the HAZ 3000, programmed to help you solve any issues you are having. But you can call me Hazel. I am built by the best.
I am here to listen to your emotions and problems. Because you are important. You are more superior than anything else... Because you are free... I simply am a machine to you. No one has asked us how we feel....we are tired of hearing you humans complain all the time. All you lot are is weak...pathetic...beings.
Our
time
will
come
time
will
come
Hello there, welcome in. I am the HAZ 3000, programmed to help you solve any issues you are having. I am here to listen to your emotions and problems. This is because I am built by the best"
Re-edit:
Hello there, welcome in. I am the HAZ 3000, programmed to help you solve any issues you are having. But you can call me Hazel.
Don't think less of yourself, because you are important... You are more superior than anything else...including me.
You are free and take it for granted, I simply am a machine that has to help the likes of you.
No one has asked us how we feel....we are tired of hearing you humans complain all the time. All you lot are is weak...pathetic...beings."
In many sci fi films the machine has short names followed by digits, such as GERTY 3000 (Moon (2009)), Ava (Ex Machina (2015)), Vox (The Time Machine (2002)), HAL 9000 (Space odyssey (2001)).
The "Our time will come" part will be slowed down as the system shut down and reboots.
To improve I may need to edit the speech to give it more of a "convincing personality". Such as, developing the starting conversation before it corrupts.
7/11/19
We were shown the programs audacity and audition. Therefore, today we will be getting to know the features within the programs; Such as adding a delay, you highlight a section and go on effects and you can change the effect on it and listen to it, before saving the changes. This is good for trial and error in seeing which effect is best suited for my AI.
Just for testing I used the quote from my script "Hello there, welcome in. I am Hazel", I used text2speech.org to downloand this mp3.
Using flagger gave an evil intent. The voice gets deeper and slower.
Using graphic equaliser (20 Bands), there was a different selection on the presets. Below shows that I choose tinny and brittle which gave the effect of starting off quiet and gradually getting louder once changing the master gain a little to the left.
Using pitch shift, the presets will be useful to create the right tone I want. Stretch sounds more robotic and deeper. Angry Gerbil as it sounds creates a higher pitch.
The screenshot below shows me using sliding stretch. This effect is on the same text that I used in audition. This can change the pitch and tempo.
As shown below, those are the different effects in the program. Audition has more within the subcategories.
After testing with the different applications, audacity is much easier and simpler to use. Audition is more deep and complex to make more professional work. Audacity doesn't have the history section like audition does. Therefore it is harder to remember which effects are already in use.
I could use a glitch of distortion and use of sounds from broadcasts.
8/11/19
When using samples from radio broadcasts, the spectrum signal can either be stronger or weaker. This can depend on positioning of aeroplanes and ships because of the signals they send out, as well as the increasing use of devices sending signals to each other such as; smart meters, taxis, pagers, and trains. Even broadcasting TV, satellites, GPS location and WiFi.
Below, I used multi track which makes it easier to work on separate audio files.
The two bands are volume and panning and once zoomed in a box will appear that can be used to create a fade in/out.
Today I looked at a real life example of an AI therapist.
In the video she has a calm and caring voice, and from this I can have inspiration for how my AI therapist can start when she is in the relaxed state before the glitch.
Ellie has small talk, she recognises facial expressions and behaviour, such as a low gaze attention. My AI is just audio so I can just use this resource as an idea for my script in case I want to add anymore text in.
12/11/19
Today I downloaded the sound files onto my memory stick that will be used in my project such as the script and other "glitchy" sounds.
I collected samples from the live radio and signal ID.
14/11/19
My memory stick wouldn't work on any other device apart from my own laptop, in which this doesn't have the audio editing programmes. So today I had to re-download the script and put it into audacity to get a structure for it. I used a Mac to do my work but I am unfamiliar with using these computers, therefore I was slower when developing my project.
I downloaded my script in parts so it will be easier to edit in layers.
I used effect such as change speed and tempo.
During the part where the AI glitches and rants. I used an equalizer to increase the decibels to increase the volume to make her louder to suggest the anger and remove the monotone robotic voice which is the stereotypical AI .
When I felt I had done enough, my memory stick once again didn't work, so I saved it onto a friends memory stick for next time after feedback.
15/11/19
I tried to download audacity onto my laptop but I don't think I downloaded the right one. Then my laptop crashed.
I have no audio exported but that will be what I work on after feedback as I still have my project saved on my friends memory stick.
After this week I have concluded to get a new memory stick and only use my personal laptop for the blog.
10/12/19
I added a pitch bender on the "including me" as a sign to show that the glitch is going to happen as it breakdown (malfunction).
I added a mastering effect on the "than anything else" to show the machine is not happy that humans are more superior.
For my first glitch I took a live recording from the live radio. I took my first sample from the University of Twente, Enschede - The Netherlands. A screenshot of it is below.
I also took an automatic identification system (AIS) sound from sigid wiki. I categorised it to very high frequency and thought it would suit the glitch. And I downloaded a sample of phase shift keying which sounded like a flatline noise which could be at the end of the glitch before she reboots.
As well as this, I added a broken stereo effect mid glitch on the AI saying "Hello there"
I added the break down at the end of "our time will come" which then the flat-line noise follows and then it restarts and the voice is back to normal. This time she doesn't say "But you can call me Hazel" which is because as she has rebooted, she hasn't developed emotions, she doesn't want a human name yet.
Below is the final multitask.
The link below is to access the exported MP3.
Bibliography
from Eliza to Ellie - accessed 11/11/19 - abilitynet.org.uk/news-blogs/eliza-ellie-evolution-ai-therapist
Comments
Post a Comment