Scripts to generate short music functional ear training samples with answers embedded in id3 tag lyrics. An example set of samples are available in a zip file, MrEar.zip here.
The Mr. Ear ear training music samples available for download were generated by the Python scripts in this repos. The scripts depend on the Petrucci melody ngram datasets available here from Vladimir Viro.
I began experimenting with these scripts when I realized through much effort at musical ear training, that the process of developing a good ear for relative pitch in tonal music was going to take many hours. I had surmised that people who had really good ears had generally spent many hours playing by ear, ie, learning to hear something and then play it, initially by trial and error, and eventually with less effort. That activity is akin to the process of transcribing, which I interpret as meaning to go directly from sound to symbol (which can either be demonstrated by notating the music or just playing or singing the music using scale degrees or solfege). I tried using ear training software, but my ability was so basic, it was too boring. I wanted to try something more akin to transcribing. I hypothesized that a large data set of short mp3 files with music samples that I could put on my mobile phone might be able to exercise the same brain functions involved in transcribing music. In order for the music samples to be useful, I knew I needed a way to be able to check my answers, and I eventually discovered that answers could be put into id3 tags within the mp3 files.
With a trained ear, it is possible to put all the samples into a single playlist and just by listening, be able to determine the pitch collection and sequence of pitches. I've worked with various collections of scales and modes, and with added chromatic notes. To train the ear with these samples, it is probably best to take each collection of samples on one at a time. Initially, I started with the Diatonic collection (Ionian mode). Since I couldn't even guesss the answer intially, I would just check the answer and sing along with solfege syllables. I restrict my solfege to just 12 syllables, do ra re me mi fa fi so le la te ti, (phonetically - doe rah ray may mee fah fee sow lay lah tay tee). The answers in the lyrics use scale degrees (1 ♭2 2 ♭3 3 4 ♯4 5 ♭6 6 ♭7 7), which to me are less confusing than written solfege syllables. The effort required to get fluent at translating between the 12 written scale degrees and the 12 spoken solfege syllables was miniscule compared to the ear training itself.
I found that just listening to samples, trying to come up with the answer before peeking, and then continuing to sing the sample for a minute or two before moving on has greatly improved my ear. I liked doing it a lot more than using ear training software, and all I needed was my phone. Having bootstrapped myself into a better set of ears, I am lately getting back into ear training software and transcribing.
Each of these samples starts with a tonic cue, which is an arpeggiated major or minor triad. This provides some context to establish a key. I have experimented with no cue (sticking with a single key) and 1,5,1 as a cue also. Adding the cue happened after I read Dan Harrison's books, which emphasized the importance of the dominant pitch for tonal hearing. Using the cue to try and force your ear into a key allows for hearing chromatic passages as outside notes. My goal has been to recognize all 12 relative pitches.
To try out some samples, find the zip file, MrEar.zip, for the time being I will just host it on my Google Drive here, download the zip file, and import the collections in the folders into a music library or onto a mobile device. Use a music app that can display id3 lyrics embedded in mp3 files (I have used the built in iPhone music app and MP3+ Player for iPhone). I would be interested in knowing of others for iPhone. Getting new samples onto the iPhone using iTunes while experimenting has been a tremendous pain! And one version of IOS dropped the lyrics display capability altogether (but now back).
I always use random shuffle and repeat 1 song when playing them.
Anyone interested in trying something different could vary the steps outlined below. Maybe try some of the options. I could say a lot more about things I've tinkered with.
The following steps assume a unix environment (I use Cygwin).
Put the imslp-interval-%dgram-20110401.csv files in ../Petrucci, cd into there and run
../motif-generator/reformatPetrucci.py 1 12
Just run:
genmotifs.py
Although there are many options, by default, it should generate two files gm.log and gm.mid
For this step I use Reaper, a Digital Audio Workstation (DAW) application. It would be nice to automate this step but I liked the sound quality I could get doing it by hand.
Just run:
clipmotifs.py gm
This should generate 9 subdirectories with 400-500 .wav samples each. The 9 subdirs contain melody samples from: Diatonic (Ionian), Mixolydian, Aeolian, Dorian, Harmonic Minor, Melodic Minor, Phrygian, Major Plus 1 Chromatic, and Minor (Aeolian) Plus 1 Chromatic.
This step also generates the gm_mp3.sh file for the next step.
Just run:
gm_mp3.sh
This generated script runs lame to convert the .mp3 and then runs the python script - tagmotifs.py
in order to extract the answers from the gm.log file and insert them as id3 tags in the .mp3 files.
The .mp3 files are written to a directory tree under 'gm_mp3'.
Added an option to generate samples of chromatic dyads (two notes from the chromatic scale, either ascending, descending, or harmonic (simultaneously struck):
genmotifs.py --dyads
genmotifs.py --dyads --descending
genmotifs.py --dyads --harmonic