2. EE2F2 - Music Technology 10. Sampling

3. Early Sampling It’s not a real orchestra, it’s a Mellotron It works by playing tape recordings of a real orchestra Each key starts the relevant tape playing When the key is released, the tape rewinds automatically The tapes are only 8 seconds long though, so notes can’t be held any longer Listen to the backing violins in this David Bowie recording from 1969.

4. Digital Sampling With modern computers, it is fairly trivial to digitally record sounds into memory (or disk) and then play them back in response to MIDI commands To create a digital version of a Mellotron you need to record the instrument(s) separately for every possible pitch. Extra desirable improvements: Extending the playback time Managing the memory/recording requirements Introducing velocity sensitivity Modifying the recordings to produce new sounds

5. Memory Requirements Early digital samplers were heavily restricted in memory For example Ensoniq Mirage (1985) 8 bit resolution 144 kbytes memory, 32 kHz sample rate 4.6 seconds recording time Mellotron (1963) 35 notes each with 8 seconds of tape Total = 35 x 8 = 280 seconds recording time How could just 4.6 seconds be any use?

6. Looping After an initial attack portion, many sounds don’t actually change that much with time Instead of sampling several seconds of a sustained note, it would save memory to just repeat (or ‘loop’) a small section Unlike tape, digital memory can be accessed randomly, so looping is fairly easy Flute sample With looping

7. Advanced Looping Looping points must be chosen with care to avoid: Big changes in the volume of the sound (creating a ‘pumping’ effect) Changes in phase (creating audible clicks) E.g. 1 E.g. 2 Example!

8. Pitch vs. Playback Rate Compare these two flute samples The waveform shapes are virtually identical The higher note can be generated by playing the lower note at a faster speed Being a semitone apart, the ratio between the speeds is 1:2 1/12 = 1:1.0595 Time Flute (F4) Flute (E4) Example!

9. Multi-Sampling If you speed up a sample: The pitch rises The entire frequency response of the instrument is effectively shifted Result: it starts to sound ‘squeaky’ If you slow down a sample: The pitch falls The frequency response changes too Often, the resulting sound lacks upper harmonics It sounds thin, hollow and simply wrong! Solution: Multi-sampling Example!

10. Velocity Variations Up to now, our sampler doesn’t respond to velocity Acoustic instruments respond in terms of: Amplitude: Higher velocity = Louder Envelope: The sound can evolve in a different way Timbre: Spectrum changes with velocity Amplitude and frequency variation could be achieved using an amplifier and a filter (more later) Alternatively, record samples of the actual sound for different velocities

11. Cross-Fading In a multi-sampled set-up, the sample used for any note can depend on: Pitch Velocity The simplest implementation is to allocate each sample a ‘zone’ in pitch/velocity space This can mean there is a sudden transition when crossing zone boundaries To prevent this, interpolation (or cross-fading) can be used Example!

12. Sampling Limitations Sampled instruments can sound very realistic but they do have some notable limitations: High memory demands It’s difficult to sample expressive instruments whose sounds can evolve in response to the performer Only real instruments can be sampled! A way of addressing some of these issues is to combine sampling and subtractive synthesis techniques Such instruments are known as sample & synthesis instruments

13. Amplitude and Timbral Control As noted before, acoustic instruments respond in terms of: Amplitude: Higher velocity = Louder Envelope: The sound can evolve in a different way Timbre: Spectrum changes with velocity All these effects can be modelled using familiar processes Amplitude changes can be synthesised using an amplifier Envelope changes can be synthesised by varying the parameters of an envelope generator Timbral changes can be synthesised using a filter All of these are found in a subtractive synthesiser

14. Sample + Synthesis This diagram is identical to the subtractive synthesiser except that the V.C.O is now a sampler It can be playing very short loops or entire recordings Multi- sample playback L.F.O. Output V.C.F.V.C.A. Env. Gen. Trigger

15. Using Sample + Synthesis Some examples of how sample+synthesis can be used: Instead of recording multiple velocity samples, just record one loud sound and model velocity effects using a filter and amplifier To apply performance effects like vibrato or crescendos, use the LFO or amplifier respectively To create entirely new sounds, use the processes in the same way as an analogue subtractive synthesiser Example!

16. Comparison with Analogue Synthesis Amongst the samples stored in memory, standard waveforms like sines, squares etc. can also be stored The sample & synthesis structure can, therefore, do everything that a subtractive synthesiser can Only notable exceptions in early models are large frequency sweeps (portamento or ‘slide’), especially when using multi-samples The latest instruments can even handle this.

17. Pros & Cons Pros Very convincing, realistic sounds are possible Compared with other techniques, it’s relatively easy to program a sampler Cons Hard to innovate and produce novel sounds – only real instruments can be sampled Very unconvincing performances are possible – some instruments don’t sample well SuitableUnsuitable Pianos Drums Ensembles Solo strings Solo brass Individual voices ‘One-shot’ sounds or relatively inexpressive Expressive sounds. Smooth transitions between notes.

18. Summary Sampling Playback of pre-recorded sounds Pitch can be varied by changing playback rate Multi-sampling improves quality over whole pitch & velocity range Sample + Synthesis All the benefits of high quality sampled sounds plus the ability to form novel instruments Easily the most popular technique currently used