Data analysis – all pages
Start
SInES Tools
Projects and systems
Data analysis › Start › Start here: from recording to result
Learning path · for Master’s and PhD students
From recording to result in eight steps
A way through a small movement or sound study with free tools – the same steps Healthy Strings uses. Each step names the tools and the mistakes we made ourselves.
1 · Question and measure
Write down which movement or sound property answers your question – and how you will measure it: a range of motion, an angle, a speed, a loudness or an intonation curve. Decide before recording which comparisons you need (players, conditions, tasks) and how many repetitions.
2 · Ethics, consent and data protection
Videos show identifiable people. Get written consent for recording, analysis and any use in teaching or publications. The SInES Tools run locally in the browser – nothing is uploaded. Cloud services (for example OpenCap) upload videos to a server: only with explicit consent.
3 · Recording
Follow the Healthy Strings recording protocol: two cameras on tripods (30° front-right and 90° right side), whole body, violin and bow in view, constant frame rate, one visible and audible clap at the start (and ideally at the end).
- Switch off mirroring in the camera app – a mirrored video swaps left and right in every pose model.
- Contrast between clothing and background; long single-coloured sleeves make the arm harder to track.
- Same phone model and settings for both cameras, or an external recorder for the sound.
4 · Prepare
Convert phone videos to mp4 and extract the sound (video tools). Check the frame rate – phones often record with a variable frame rate. Cut the recording into tasks, by hand or with the sound level (onsets, clap).
5 · Extract
Run the pose tracker on each video and the sound tools on each audio file. Write down tool, model, version and every setting – you will need them for the methods section.
6 · Check quality
Look at the result next to the video (3D Data Analyzer or stick figure) before computing anything.
- Segment lengths (e.g. upper arm) should stay nearly constant – a large variation means tracking noise.
- Jumps, swapped sides, hidden body parts (the left arm in the side view).
- Depth (z) and model-based 3D are estimates; compare cameras before trusting them.
7 · Merge and compute measures
Bring pose, sound and ratings onto one time axis (Excel Synchroniser), using the clap to correct the offset between cameras. Compute your measures per task – for example the range of motion (RMS of the displacement), joint angles or speeds. Movement-quality features: PyEyesWeb.
8 · Statistics and reporting
Explore in the browser, analyse in R or JASP: repeated measures and mixed models, agreement between raters (ICC), corrections for multiple tests. Report tools, versions and settings and cite them properly. Remember what the tools cannot do: video describes movement – it does not show forces, muscle activity or injury risk.
Common mistakes
| Problem | Solution |
|---|---|
| German export with semicolon and decimal comma | R read.csv2(), Python sep=";", decimal=",", or the CSV Formatter |
| Left and right swapped | mirroring off; name sides functionally (bow arm, violin arm) |
| Fast movements look small | tools smooth the data (about 0.1 s) – vibrato and tremolo are damped |
| Time column does not start at 0 | always use the time column, never frame number ÷ frame rate |
| Cut with “copy” is off by a second | re-encode when cutting; check variable frame rate first |
| Fixed seconds grid fails | players skip lines or rest longer – check every segment in a plot |
These points come from the Healthy Strings pilot studies. Questions: Healthy-Strings@matthias-bertsch.at.