Transcribing piano from recording when available recordings are poor

Question

I do a fairly large amount of transcribing of various jazz material. Usually, I do transcriptions of solos with only vague concern for what the accompaniment is doing (since usually the harmony is a known quantity). Recently, however, I have shifted my focus towards transcribing solo piano passages.

My question is this: What can you do to make a good and accurate transcription when there are simply no crisp, clear recordings and no reference transcription exists, particularly when the object of the transcription is the piano? I consider this to be a question about tools in the literal sense (software and whatnot) and to be about the kind of knowledge that one would utilize in transcribing piano.

To add some context: When transcribing melodic passages, I usually don't use any tools; I might slow the recording down, but that's about it. When you introduce multiple notes, things get hairier, and I have at times employed spectrograms (like those produced by AnthemScore) if I really can't figure something out, but their value is somewhat questionable, particularly for live recordings. I am also aware that, as a person who does a lot of transcribing, logic sometimes rules the day over ears and you employ what you know about music to fill in what you can't hear. However, on some of these recordings the quality is so poor that of voicings I don't hear much of anything at all. I could tell you the chord, of course, but not how the player actually played it, which is often the main purpose of the transcription itself.

I am particularly looking at transcribing Ahmad Jamal's famous performance of Poinciana that appears on At the Pershing: But Not For Me, for which some transcriptions of dubious quality exist but none seem to be especially good. This is, however, a general question, and the example I'm providing here is simply a reference point for the kind of work that I'm trying to do.

Answer 1

I don't think that you're going to find very satisfying answers, mostly because I am fairly confident that none exist. That said, our brains are amazing at pattern matching, particularly when coupled with expertise. Our brains see patterns where there aren't any, and can make sense of patterns of behavior after a single experience. We can identify probable mistakes with quite a lot of accuracy, and generally find information from rather incomplete information.

To maximize these properties, I have these recommendations:

1. Give yourself a lot of exposure. Give yourself a lot more time than you usually would for a transcription. Listen a lot more than you usually would. Give your brain time to make all of those connections and hear all of the subtle clues that might influence the choices you make in voicings. The more time you give this, the more like you are to not miss the subtle differences.
2. Zoom out before you zoom in. Some of our best pattern-matching work is done from a distance. Listen holistically, and trust that your mind will find the sound groupings that will ultimately help you find reasonable voicings.
3. Embrace guesswork. You are already an expert thinker in this field. Let logic and intuition guide you where your ears cannot.
4. Play through the transcriptions that already exist. Sit with the recordings, and play through the transcriptions that you have. In the spots where you are unconvinced by what they have done, see if you can cobble something that is more accurate together.
5. Accept imperfection. At the end of the day, if the information isn't there, it isn't there. On a really low quality recording, there will be things that you can't ever get quite right, and you'll ultimately have to be okay with that.

Answer 2

Ways to Improve a Recording

Something you can try is importing the song into audio editor software like Audacity and modifying the recording to highlight the piano. If it's a stereo recording, you can isolate the channel in which piano is more prevalent and increase that channel volume to make the piano louder relative to the band. Another thing you can try is to adjust the equalizer settings to highlight the piano frequencies. GarageBand has equalizer presets and adjustment tools specifically for things like this. This could entail listening to the recording through GarageBand, watching the graph for likely piano frequencies, and increasing those piano frequencies in order to highlight the chord voicing. Finding the right equalizer settings likely will involve some guesswork, and it obviously can change based on things like recording quality, range of voicing, and piano being used. I haven't done this enough to offer guiding principles for piano transcriptions specifically, but perhaps others will chime in.

Part of this equalizer work can be done in tandem with frequency graphs from other parts of the song. If you listen to a clip of a drum solo and identify the higher frequencies that are simply produced by the cymbals, then you can reduce the levels on those frequencies to emphasize just the piano.

If All Else Fails

Sometimes there is no way to improve the quality of the recording, and you're simply stuck with a scenario where the certain notes cannot be obtained from the piano. Here are the tricks I usually try (some of which you've already described) to fill in the gaps when dealing with a low-quality recording. First, I look to see if the "holes" (notes that were played but were inaudible) fit within any broader patterns. This can apply to transcribing a melody/solo line--for example, I'll look for arpeggios, scale runs, and scale patterns that seem to be missing a note that was likely played but simply cannot be heard on the low-quality recording. This technique can also extend to transcribing chords--sometimes there are obvious-enough gaps in voice leading, inner voice movement, upper structure triads, etc. to infer that certain notes that were likely played but not audible.

These are some of the ways to follow logic, as you nicely put it. But if this fails, then in my experience, the best next thing is to become extremely familiar with that pianist's playing. My first big transcription project was to transcribe all of Kenny Barron's album Green Chimneys. Halfway through the project, I could somewhat regularly predict the end of a solo line/lick after hearing only the beginning of the line/lick. Intuition was kicking in--and this intuition was owed to the other transcriptions I had done of other solos earlier on the same album. (I was only transcribing the solo lines.) But I've found something similar to be true with Chick Corea voicings and Bill Evans voicings (though to a lesser extent because I've spent less time transcribing their chord voicings).

Unfortunately, most people don't have the luxury of transcribing 5 other songs purely for the purpose of aiding an original transcription with inaudible/indistinct notes. So, a more direct route is to find other versions of the same song performed by that same pianist. As you'll appreciate, some pianists are fairly fixed in their solo arrangements, and from one recording to the next, there won't be much deviation in the notes being played. Listening to other live versions or studio versions of the same solo piano piece can offer insight into the hard-to-hear (or impossible-to-hear) notes that have been lost to a low-quality recording.

This is an effective tool for voicings. My experience has been that voicings are re-used even more than solo lines. If you know of a second recording of the same song by the same pianist, you can start by limiting your guesswork to voicings that appear in the higher-quality recording. In many instances, there will be a match. This type of matching greatly reduces the number of possible voicings you're considering for a given chord. For example, if you know that an A7sus chord is being played, but you can't determine the specific notes, you can start by looking at the ways the pianist voices A7sus on the other recording, and check for a likely match. Even if you're not doing this for every chord, this can still be a tedious process, but over time the voicings become internalized, the same way that one can internalize improvisational lines.

Using Poinciana as an example, you might try transcribing one or more of these recordings (example 1, example 2, example 3, example 4) where Ahmad Jamal is playing the same song but at a different concert. At first glance, it sounds like his chord voicings over the melody are mostly the same, even in the more recent concert that occurred decase after the '58 recording. There won't be this same 1-to-1 match in the voicings he uses when soloing, but I bet there is still a lot of overlap.

Ongoing Ear-Training

This may not apply to you so much as other readers, but it will be equally useful to focus on continued ear-training. The process described above can work in moderation, when dealing with a few chords here or there that simply are not discernible from the raw recording. But trying to engage in this process for every chord in a recording is probably not manageable. If that's the scenario one finds oneself in, then either the recording is un-salvageable or more ear-training work is needed. When transcribing chords, it helps to be able to identify all of the notes in a familiar chord when played clearly. If one isn't able to recognize all of the notes in a rootless C7#9b13 voicing, then it's going to be even more difficult to do this with a less-than-perfect recording. Perfecting that skill (through ongoing ear-training) makes a big difference when tackling poor-quality recordings.

Answer 3

You persist. Use a looping and slow-downing tool like Transcribe! Gradually you will get more and more of it right. Experience of the style helps, but don't fall into the trap of writing what you think SHOULD have been played - maybe he was being original! Sometimes it's easier to detect what notes AREN'T being played than what notes ARE. So what's left must be right :-)