Question

Let me first establish what I want to do:

My user is able to record voicenotes on my website, add tags to said notes for indexing as well as a title. When the note is saved I save the path of the note along with the other info in my DB.

Now, I have 2 choices to do the recording, both involve a .swf embedded in my site:

1) I could use Red5 server to stream the audio to my server and save the file and return the path to said file to my app to do the DB saving, seems rather complicated since I would have to convert the audio and move it to the appropriate folder that belongs to the user in a server side Red5 app, which I’m not very aware of how to build.

2) I could simply record the audio and grab its byte array, do a Base64 encoding on it and send it to PHP along with the rest of the data that is necessary (be it by a simple POST or an AJAX call), decode it on the server and make the file with the appropriate extension, audio conversion would also occur here using ffmpeg, this option seems simpler but I do not know how viable it is.

What option would you say is more viable and easier to develop? Thanks in advance

Answer 1

Depending on the planned duration of the recording, you may very well be able to use option number two. I recently used a similar approach successfully for a project, but recordings were only up to 30 seconds or so. Here’s what I did differently from what you’re suggesting though, and why I think it’s better:

• To capture the sound from the microphone and store it to a ByteArray, use the SAMPLE_DATA event which is dispatched whenever more sound data comes in from the microphone. There’s an example in the documentation that should explain this well enough.

• Because most users would be on normal home computers without any special recording equipment, it was safe to assume that the full fidelity of the recording is not necessary. I used just 2 bytes per sample, and only mono, instead of using the full 64 bit floats (AS3 Number) that you get from the microphone on the SAMPLE_DATA event. Simply read the Number and do myFloatSample * 0x7fff to convert to 16 bit signed integer.

• Don’t use the native 44.1kHz sampling rate if you’re just recording speech or something else in that frequency range. You will likely get away just fine with 22.05kHz, which will cut the amount of data in half straight away. Just set the Microphone.rate property accordingly.

• Don’t use Base64 to encode your data. Send it as binary data, which will be significantly smaller. You can send it as raw POST data, or using something like AMF. Also, before you send it, use the native compress() or deflate() methods on the ByteArray to compress it. On the server, decompress using the ZLIB or raw DEFLATE (inflate) algorithms respectively, which PHP supports.

• Once decompressed on the server, what you have is essentially what is called a raw 16-bit mono PCM stream. Incidentally, that should be one of the very input formats that ffmpeg (or lame) supports, so you should be able to encode it to mp3 without having to do any manual decoding first.

Obviously the Red5 solution will likely be better, because it’s more tailored for the task. But if you don’t have the resources to set up a Red5 server, or don’t want to use Java, the above solution is proven to work well as long as you stay away from too long recordings.

To take a simple example, a 30 second recording at 22,050 samples per second, 2 bytes per sample will be ~1.3MB. Even once deflated, the transfer to the server will likely still be almost a megabyte for 30 seconds of audio. This may or may not be acceptable for your application.