taps2beats is a somewhat (?) oddball command-line utility and Go module that estimates the beats of a piece
of music from a file of the beats as tapped by a person (or other musical entity of whatever sort).
The internal algorithm uses an implementation of Ckmeans.1d.dp to cluster the supplied 'taps' into the optimal equivalent beats, followed by least squares regression to estimate the BPM and offset and (optionally) quantize and (optionally) interpolate the beats over the interval.
Goversion 1.15+
To install the stable version, download the tar.gz or zip release for your platform from the Releases
page, and unpack into the directory of your choice.
The build the development version, clone the taps2beats github repository:
git clone https://github.com/transcriptaze/taps2beats.git
cd taps2beats
make build
taps2beats [options] <file>
The file is a list of lines of the 'taps' (in seconds), separated by whitespace e.g.
4.57027 5.063594027 5.6035973 6.1026998 6.642708943 7.14179696 7.7107 8.1470916
4.52956007 5.069284016 5.603428973 6.102591998 6.613455 7.14764 7.69912088 8.215609871
4.517865093 5.022782107 5.580101018 6.096715 6.654118921 7.1763719 7.681405914 8.215537871
5.1392891 5.545395086 6.067721066 6.578564068 7.1300991 7.65971 8.134273
4.4981138 5.040234073 5.562732052 6.079333043 6.624973977 7.14165 7.66 8.19905
4.511093 5.069403016 5.586174007 6.108568986 6.57864 7.147523 7.681614 8.262078
There is no requirement that the number of taps on each line be the same or that the taps are specified to the same precision. Ideally, each line represents a loop of the music being tapped, but this is only significant if a forgetting factor (described below) is used to weight later 'taps' as being more accurate than e.g. the first few attempts.
If the input filename ends with '.json', the file is parsed as a JSON object that is expected to contain:
{
"taps": [][]float
}
Invoking taps2beats without an input file reads the taps from stdin.
The output format is a fixed column width list of beats, with each beat represented by a line that contains
- Beat number
- Estimated beat
- Average of the 'taps' for the beat
- Variance of the 'taps' for the beat
- The 'taps' from the input file used to estimate the beat
| Beat | At | Mean | Variance | Taps |
|---|---|---|---|---|
| ... | ||||
| 6 | 7.153s | 7.153s | 1.3ms | 7.142s 7.136s 7.177s 7.148s 7.176s 7.13s 7.13s 7.165s 7.148s |
| 7 | 7.686s | 7.686s | 2.1ms | 7.711s 7.693s 7.699s 7.681s 7.652s 7.722s 7.653s 7.687s 7.682s |
| ... |
Options:
taps2beats [--verbose] [--out <file>] [--interval <interval>] [--quantize] [--forgetting <factor>] [--precision <time>] [--latency <time>] [--shift] <file>
--verbose Displays operational information
--out <file> Writes the estimated beats to the supplied file
--interval <interval> Extrapolates (and interpolates) the beats to extend over
the supplied interval. The interval should be specified as
<start>:<end> where <start> and <end> are in Go time format
e.g. --interval 0.3s:1m10.3s.
A '*' interval (--interval '*') will interpolate the beats
over the interval from the earliest to the latest 'tap'.
--quantize Adjusts the estimated beats so that they fit to the estimated BPM
--forgetting <factor> Discounts earlier taps from earlier loops as being less accurate
than later loops due to the listener learning the music. e.g. a
factor of 0.1 discounts each loop by 10% over the subsequent one.
A negative factor inverts the weighting i.e. earlier loops are
weighted as more accurate than later loops.
--precision <time> Rounds the beats and all times to the specified precision (in Go
time format) e.g. --precision 1ms will round all times to the
nearest millisecond. The default precision is 1ms.
--latency <time> Adjusts all times to compensate for the latency between the
actual beat and the detected 'tap' e.g. --latency 73ms
--shift Adjusts all beats (and times) so that the first beat in the
interval falls on 0s.
--json Formats the output as prettified JSON, with all the times converted
to seconds (to a precision of 1ms)
taps2beats examples/taps.txt
BPM: 114
Offset: 316ms
1 4.524s 4.524s 0s 4.57s 4.506s 4.53s 4.53s 4.518s 4.495s 4.529s 4.524s 4.518s 4.518s
2 5.058s 5.058s 1ms 5.064s 5.046s 5.058s 5.069s 5.023s 5.133s 5.04s 5.04s 5.046s 5.046s 5.069s
3 5.578s 5.578s 0s 5.604s 5.592s 5.592s 5.603s 5.58s 5.545s 5.563s 5.557s 5.586s 5.551s 5.586s
4 6.1s 6.1s 0s 6.103s 6.114s 6.137s 6.103s 6.097s 6.068s 6.079s 6.132s 6.091s 6.074s 6.109s
5 6.618s 6.618s 1ms 6.643s 6.619s 6.631s 6.613s 6.654s 6.579s 6.625s 6.654s 6.596s 6.608s 6.579s
6 7.153s 7.153s 0s 7.142s 7.136s 7.177s 7.148s 7.176s 7.13s 7.142s 7.194s 7.13s 7.165s 7.148s
7 7.686s 7.686s 1ms 7.711s 7.693s 7.699s 7.699s 7.681s 7.652s 7.664s 7.722s 7.653s 7.687s 7.682s
8 8.21s 8.21s 1ms 8.192s 8.204s 8.227s 8.216s 8.216s 8.134s 8.198s 8.245s 8.181s 8.239s 8.262s
./bin/taps2beats --verbose --precision 1ms --latency 7ms --quantize --interval 1s:12s examples/taps.txt
taps2beats v0.1.0
... reading data from ./runtime/taps.txt
... 12 values read from ./runtime/taps.txt
... using forgetting factor 0
... quantizing tapped beats to match estimated BPM
... interpolating missing beats over interval 1s..12s
... 21 beats
... compensating for 7ms latency
... rounding to 10ms
BPM: 114
Offset: 310ms
1 1.36s
2 1.89s
3 2.41s
4 2.94s
5 3.47s
6 3.99s
7 4.52s 4.52s 0s 4.56s 4.5s 4.52s 4.52s 4.51s 4.49s 4.52s 4.52s 4.51s 4.51s
8 5.04s 5.05s 0s 5.06s 5.04s 5.05s 5.06s 5.02s 5.13s 5.03s 5.03s 5.04s 5.04s 5.06s
9 5.57s 5.57s 0s 5.6s 5.58s 5.58s 5.6s 5.57s 5.54s 5.56s 5.55s 5.58s 5.54s 5.58s
10 6.1s 6.09s 0s 6.1s 6.11s 6.13s 6.1s 6.09s 6.06s 6.07s 6.12s 6.08s 6.07s 6.1s
11 6.62s 6.61s 0s 6.64s 6.61s 6.62s 6.61s 6.65s 6.57s 6.62s 6.65s 6.59s 6.6s 6.57s
12 7.15s 7.15s 0s 7.13s 7.13s 7.17s 7.14s 7.17s 7.12s 7.13s 7.19s 7.12s 7.16s 7.14s
13 7.67s 7.68s 0s 7.7s 7.69s 7.69s 7.69s 7.67s 7.65s 7.66s 7.72s 7.65s 7.68s 7.67s
14 8.2s 8.2s 0s 8.19s 8.2s 8.22s 8.21s 8.21s 8.13s 8.19s 8.24s 8.17s 8.23s 8.26s
15 8.73s
16 9.25s
17 9.78s
18 10.3s
19 10.83s
20 11.36s
21 11.88s
taps2beats --json examples/taps.txt
{ "BPM": 114, "offset": 316000000, "beats": [ { "at": 4524000000, "mean": 4524000000, "variance": 0, "taps": [ 4570000000, 4506000000, ..., 4518000000 ] }, { "at": 5058000000, "mean": 5058000000, "variance": 1000000, "taps": [ 5064000000, ..., 5069000000 ] }, ... { "at": 8210000000, "mean": 8210000000, "variance": 1000000, "taps": [ 8192000000, ..., 8262000000 ] } ] }
BPM: 114
Offset: 316ms
1 4.524s 4.524s 0s 4.57s 4.506s 4.53s 4.53s 4.518s 4.495s 4.529s 4.524s 4.518s 4.518s
2 5.058s 5.058s 1ms 5.064s 5.046s 5.058s 5.069s 5.023s 5.133s 5.04s 5.04s 5.046s 5.046s 5.069s
3 5.578s 5.578s 0s 5.604s 5.592s 5.592s 5.603s 5.58s 5.545s 5.563s 5.557s 5.586s 5.551s 5.586s
4 6.1s 6.1s 0s 6.103s 6.114s 6.137s 6.103s 6.097s 6.068s 6.079s 6.132s 6.091s 6.074s 6.109s
5 6.618s 6.618s 1ms 6.643s 6.619s 6.631s 6.613s 6.654s 6.579s 6.625s 6.654s 6.596s 6.608s 6.579s
6 7.153s 7.153s 0s 7.142s 7.136s 7.177s 7.148s 7.176s 7.13s 7.142s 7.194s 7.13s 7.165s 7.148s
7 7.686s 7.686s 1ms 7.711s 7.693s 7.699s 7.699s 7.681s 7.652s 7.664s 7.722s 7.653s 7.687s 7.682s
8 8.21s 8.21s 1ms 8.192s 8.204s 8.227s 8.216s 8.216s 8.134s 8.198s 8.245s 8.181s 8.239s 8.262s
Please create an issue in the taps2beats github repository.
-
Wang H, Song M (2011). “Ckmeans.1d.dp: Optimal $k$-means clustering in one dimension by dynamic programming.” The R Journal, 3(2), 29–33. doi: 10.32614/RJ-2011-015.
-
Song M, Zhong H (2020). “Efficient weighted univariate clustering maps outstanding dysregulated genomic zones in human cancers.” Bioinformatics. doi: 10.1093/bioinformatics/btaa613, [Published online ahead of print, 2020 Jul 3].
-
Zhong H (2019). Model-free Gene-to-zone Network Inference of Molecular Mechanisms in Biology. Ph.D. thesis, Department of Computer Science, New Mexico State University, Las Cruces, NM, USA.
- Icons by FontAwesome: