New Version of "Blumine"

A new version of Mahler's "Blumine" (version 1.4) is now online on the VPO website (MP3 and WMA file). This version is based on the most recent rendition with trumpeter Julius Eiweck. The changes only concern the string part. Mahler-expert Jim Zychowicz had commented that the strings would benefit from "more depth", and so I decided to add a few layers. In particular the beginning sounded a bit "tinny", and so I added a total of four layers for the violins: for violin 1 a separate additional tremolo track (the GPO4 violin tremolo is different from the tremolo of the keyswitch violins) and a muted sordino, and the same for violin 2.

These changes give the bright strings a bit more substance and bite, and I find that this has improved the sound, away from the thin chambermusic-like texture to a more Mahlerian orchestra. I am working now on the same effect for all the other movements of Mahler's Symphony No.1.

Problem with Clarinets

The rendition of the first movement of Mahler's Symphony No. 1 is almost completed: missing are only the flutes, oboe, and clarinet voices from bar 263 till the end. But now I hit a slight obstacle: the Es-Clarinet requires a high A and high B to be played - however, the Garritan Personal Orchestra (GPO4) samples only go up to the high G... I have hit these limitations of the GPO4 already in other instruments earlier, when I recorded other movements, for example when some very low notes of brass instruments were not there. Mahler just uses the full range of the instruments. For some reason, GPO4 did not record those extreme notes... it would be nice if in such a case the missing notes would just be automatically extended, based on the last available sample. It would not matter if this would not sound perfect - since it is an extreme note, it is anyway not supposed to sound perfect, even with a real instrument - these extreme notes are often only used to create an effect.

I had solved this problem in other renditions by using a standard software synthesizer in SONAR: Cakewalk TTS-1. However, as soon as I loaded that synth and played the first note on it, SONAR froze, the audio engine crashed, and I had to reboot the system. This happened several times... I suspect this has to do with the fact that I have loaded many (more than 30) GPO4 samples into memory, using 4 different ARIA VST instances. Should theoretically not matter, as the 12 GB memory of MAESTRO-2 are not fully used yet, but it appears that the memory management has some difficulties handling both the GPO4 and the TTS1 simultaneously.

So it looks as if I will for now not be able to play those high notes in the Es clarinet... fortunately this is in places towards the end where the music is very loud, most instruments of the orchestra are playing ff, and that single clarinet would not have been audible anyway, would only have contributed to the overall sound colour. But still, is a pity! I will have to find a solution for this eventually.

Progress on next Mahler Symphony rendition

I have started to work on the rendition of the 2nd movement of Gustav Mahler's Symphony #1. This time I applied a new method: I first played "live" on piano the whole movement, with excerpts from the part, in order to get the desired tempo that I felt would be right. I also hoped that by this method I would be able to incorporate those slight tempo variations within a theme, a motif, a phrase, and within a bar. So after the recording was completed (it took me about 3 weeks to get this initial recording done, because I often stopped and repeated, as I felt the tempo was not right), I recorded another track, which just had the beat indicators in it. I listened to my recording and just hit a key on the keyboard, linked to a percussion instrument. This track was then the reference track for the tempo, and I used the SONAR tool "Fit to Improvisation" to create the tempo map which creates a list of tempo values at every quarter note. When looking at this tempo map, I saw very interesting artefacts: there appeared some patterns in the tempo within the bars, which were not consistent throughout the piece. In some cases there was a slow-fast-slow bar, and sometimes there was a fast-slow-fast bar. I was not able to determine first if that was just through my own inaccurate playing or tempo mapping, or if there was actually some musical principle behind this. But I noticed overall that I had started quite slow, much less than was noted in the score (score: 3/4=66, mine: 3/4=50), but then the tempo accelerated throughout the movement. Not very good actually - there was no good reason for that other than I had been carried away... I now can admire the conductors' ability to keep the tempo constant.

After having the tempo map I began recording the actual music instrumental tracks. And here it became apparent that this pre-recorded tempo map was pretty useless in many cases: the tempo variations within a bar prevented me to get a constant beat where the rhythm had to be exact and quantised. This was important for instruments which established the base rhythm, for example in percussion parts, or when a set of eighth notes were to be played: these need to be exactly quantised in many cases, otherwise it sounds uneven, as if the player cannot play properly. The varying tempo that I had incorporated into my live recording, came from an interpretation of various instrumental parts which could overlay their own more "free" (rubato) voice onto that fixed and constant tempo. Now, in order to achieve this properly I had to abandon the original tempo variations, and I flattened the tempo map to mostly constant segments. Then I recorded each instrument and applied a slightly different treatment of the tempo: the important rhythm-establishing parts are quantised, whereas other parts I played the instruments freely, not exactly observing the beat times. This added a bit or realistic and more musical flow to the piece. But there I encountered the difficulty that the audio lag (due to buffer requirements) was about 30 ms (this was the lag between me pressing a note on the keyboard and then hearing it). So it was almost impossible to create an appropriate consonant accompaniment with already existing tracks in live recording, as my playing was recorded trailing behind. I could shift this after the recording, but during the play I did not really get the right acoustic feedback from my own playing; therefore I was not able to properly play the instrument tracks live but had to do a lot of manipulation and correction afterwards. I often ended up quantising whole segments, so that they matched the existing rhythm.

What might be done next time: I could set the buffer to a low value, for fast response and short lag, and then record only track by track, muting the other instruments, and pretending I would play a solo instrument. This would prevent the overload of the buffer (which occurred several times while I had 4 ARIA VST instances for all the instruments), but it would also prevent me to hear the other instruments while playing "solo": in the end the coherence may not be very good... I mayjust have to try it out to see if this is feasible.

In the meantime I reached already marker 13 (about 3 minutes) in the 2nd movement of this Symphony, with all instruments so far. It sounds a bit "tinny", more like chamber music, because I did nothing to add more "lushness". But this has also an advantage: the instrumental voices come out quite clearly, more clearly than in the traditional orchestra recordings I know. And someone had made already the same observation in the 3rd movement which I had posted a few weeks ago: the GPO samples as I have applied them do have a chamber-orchestra feel to them. I might have to double the strings and layer them with more players, to populate them and create a kind of "wall of sound" :) .

Gustav Mahler, Symphony No.1, 3rd Mvmnt (first "beta recording")

Yesterday night I have completed the first "rough" recording of the 3rd movement of Gustav Mahler's Symphony No.1. This is the first music recorded on the new MAESTRO-2 computer. It is still quite unbalanced; I need to revise some of the instrumental timbres, attack, volume balance. But I decided to put this one out on the web, as a "beta version".

I had produced the very first MIDI "rendition" of this piece back in January 1995. It was one of my first renditions, after Borodin's "Polovtsian Dances" and Smetana's "Moldau". At that time I was still limited by the 16 MIDI tracks of a standard MIDI file, and I had to do lots of tricks to create a reasonable rendition, using the same tracks for different instruments by patch changing etc. A revision in 1998 allowed for more freedom, since I did no longer aim at creating a single 16-track MIDI file but used both my Yamaha MU-80 synthesizer and PC-based softfonts to create an MP3 recording. However, I was never happy with the result - this was just not the right sound, and I always had in mind to revise this rendition. In 2005 I started an attempt to revise this recording by including the Garritan Personal Orchestra sounds, but I only managed to get a few instruments converted: problems with an upgrade of the MAESTRO-1 PC prevented me to continue further. That PC simply could not handle the many audio and synth tracks.

Now after the new MAESTRO-2 has been set up, with the latest versions of Garritan Personal Orchestra (GPO 4), Sonar 8.5, the RAID-HD array, and the 12 GB triple-channel memory, I felt that the resources would now be adequate to tackle this piece again. Since the big Gustav-Mahler-Year is coming up (July 2010: 150th birthday; May 2011: 100th death day), I wanted to create something nice for this.

I used 5 instances of the VST synth ARIA, each containing a group of instruments. Unfortunately, the GPO4 instruments do not all cover the extreme notes required by Mahler's music, so I had to add a few tracks of standard Cakewalk TTS-1 instruments for the odd note here and there.

The revision took more work than I originally had envisioned: I had to "declutter" my tracks which still had left in them some patch changes from the original 1995 version, and clean this out. Also, because in standard MIDI the multiple playing of the same note causes phase effects, I had originally "simplified" the score by eliminating those duplicate notes (for example when the first and 2nd violins play the same note). Now, with actually having separate sound fonts available for 1st and 2nd violin, I had to recreate those duplicates again.

Yesterday night I did a "bounce-to-track" and created a WMA file, which can be accessed on my Virtual Philharmonic Orchestra website.

I still need to revise several things:
- the overall sound balance is not ideal.
- the brass sounds "too metallic" for my taste, needs to be softer.
- some individual notes are too loud, some phrasing (attack) needs to be corrected and equalised.

I could wait and post only the very final finished version, but I thought it would be nicer to show the progression of this recording in public. So I will release once in a while improved version of this recording, following my old VPO paradigm of the "rehearsal room". When the rendition is such that I feel it has reached its final state, I will then move it into the "Concert Hall".