The Bird Song Project

1996 -2005

May 23, 2005





Here are the current recordings  updated at 7:00 PM (EST) every day


In the spring of 1996 I installed in the western Catskills of New York State, a computer that was programmed to record sounds of a selected frequency range, analyze them, and save them as WAV files if they met certain criteria. The program also summarized the sounds in a format that I devised so that I could use a modem to download summaries of the days recording. The frequencies chosen to be saved were those I thought would capture the songs of Warblers that I knew were present in the area.

The computer was automatically turned on every Monday, Wednesday and Friday at 5:30 AM and was usually programmed to turn off at 8:00 PM the same day (2PM in winter)

In the course of time I would try to identify the birds on the recordings, but this turned out to be a very time consuming task and I devised a method of doing a quick estimate of what the possible species were. This method consisted of creating a frequency/ frequency modulation matrix to which I compared each summary. This first estimated was added to the record of the recording before it was saved.

The hardware and software limitations led to a cycle time of five minutes. Thus every five minutes the computer would record 10 seconds of sound, analyze it, and if it met the criteria, save both the sound (as a WAV file), an the data record to the computer.


The computer installed to monitor the sounds was, in 1996, an old piece of technology: a 80386 PC running at 20 MHz. The CPU had 4 Mbytes of RAM and the hard disk was only 500 Mbytes.The soundcard was one that is limited to recording at a sample rate of 14,925 samples/sec. The microphone was built from a simple condenser microphone element mounted in a 15-inch parabolic reflector molded from epoxy and fiberglass.

The soundcard was one that is limited to recording at a sample rate of 14,925 samples/sec. The microphone was built from a simple condenser microphone element mounted in a 15-inch parabolic reflector molded from epoxy and fiberglass.

The current hardware is a 166 MHz Pentium with two shock mounted Panasonic WM-61B omni capsules shock mounted on the corner of my house. They feed into a  Sony Pro stereo cassette recorder that supplies preamplification and Plug in Power. The output of the recorder is sent to a graphic equalizer that is set to reduce frequencies below 1KHz. The GE sends its output to the line input of my sound card. The soundcard records to .WAV files at a sample rate of 22,000 samples/sec.


The original software used to record the bird songs was written in Turbo Pascal and the routine was controlled with MSDOS batch files. This combination has proved to be very stable, and the system has run unattended for weeks at a time. Two years ago (2002) I upgraded my capture computer and operating system. I now use Windows and have rewritten the the software as a single Windows program in the computer language Delphi. The program combines the original recording and analysis programs of the original system.

Use this link to see a flowchart of the recording cycle.

Here is the log of current recordings. This file is updated hourly. The program is currently searching for approximately 20 species.



The site on which the recording was done is in a clearing on the southeast side of a mountain, at an altitude of 2,000 feet. The environment can best be described as Northern Hardwood Forest.

The predominant trees are Sugar Maple, American Beech, and Mountain Ash. The understory consists of : Striped Maple, and Blackberry.

The common birds in the area are:

Eastern Phoebe

Scarlet Tanager

Red-Eyed Vireo

Black-capped Chickadee

Yellow-Bellied Sapsucker

Dark-eyed Junco

American Robin

Rose-Breasted Grosbeak


Hermit Thrush

Black-Throated Blue Warbler

Black-Throated Green Warbler

Mourning Warbler

Chestnut-Sided Warbler

Pileated Woodpecker

Downy Woodpecker

American Kestrel

Wild Turkey

Roughed Grouse

American Goldfinch

House Finch

Purple Finch

Song Sparrow

American Field Sparrow

Indigo Bunting

White-Eyed Vireo





In order to review the large amount of data being collected I designed a database to summarize the sound file and to establish a base from which to try to identify the bird songs recorded.

The program takes an FFT of the WAV file, and then from that analyses makes a record that includes time, frequency and frequency modulation information. The record is in a text file format.

The file extension used for the data is *. DAT. Each record of the file starts with a non space character and begins with the name of the sound record. All lines following this line that start with a space are part of the same record.

In the same directory is saved the actual WAV sound file from which the record was created. If the data file was as shown below, there would also be three WAV files ( 094340.WAV, 102119.WAV, 112546.WAV ).

In some cases the WAV file will not be present. This may be because it was removed for some reason, or that the record was transferred from another file, and the WAV file was not moved.

094340 037Fi-10 118 037Fa-21 180 037JA-21 182 038FA-17 288 037Hb-21 211 


; SmpRT: 14925 Db: 10.0 MinWid:3 FilW:10 Lrgst:187.5 LrgInRng:187.5

;Andes 8/2/1996 9:43:40

102119 040Ee-15 194 034Fh-07 221 037dJ-15 211 040A--27 106 040Ig-10 089


; SmpRT: 14925 Db: 10.0 MinWid:3 FilW:10 Lrgst: 71.6 LrgInRng: 23.2

;Andes 8/2/1996 10:21:19

112546 037J--14 012 025Kd-05

; SmpRT: 14925 Db: 10.0 MinWid:3 FilW:10 Lrgst: 64.4 LrgInRng: 25.6

;Andes 8/2/1996 11:25:46


Each record in the database has several fields:

Record ID This is the name of the sound file summarized. In my recordings it is the time of the recording (hhmmss).

Note This is made up of eight(8) characters: fffmmmtt

fff is the frequency (1/s) of the note at its highest amplitude divided by 100. Thus a code of 035 encodes 3,500 Hz.

mmm is the frequency modulation (FM) of the note. Letters (A) to (P) indicate an increasing frequency and letters (a) to (p) indicate a decreasing frequency. The higher in the alphabet , the greater the absolute rate of frequency change. The largest number of inflections of the frequency is two(2), thus only three(3) different frequencies are allowed for. A (-) character indicates that there is no inflection at that point. A code of ( A--) would indicated a note of a small positive frequency change and no inflection. A code of (Ag-) would indicate a note with one inflection, starting with a small positive frequency change followed by a large negative frequency change. Click here for a more technical explanation of the FM codes.

tt is the length of the note in centiseconds(sec-2).

Silence This indicates the time between note. It is made up of three characters (sss) which encode the time in centiseconds(sec-2). The silence code must be preceded and followed by a space character.

The number of note and silence fields are unlimited as long as their is a silence field between each note field.

This notation is enough to give a crude reconstruction of the sonogram of the sound. It also has the property of being both people and computer readable. Since the database is made up of ASCII characters it is searchable by any simple word search utility as is commonly found installed on computers.

Comments These are preceded by the (;) character and may contain any information about the recording, environmental parameters, or personal observations.



Here is a flowchart of the recording sequence:

Here is an example of how I use the song database to compile matrix filters that are used for identifying songs that have been saved as a database entry.

This matrix is one that would be used to test whether the notes of an Indigo Bunting song had been detected in the note records.


For more information about the project  contact me by e-mail