Translating Animal Voice into Different languages

Really quite simple:

1. Get hardware that can capture and process both the infra-sonic and ultra-sonic frequency content that covers the range of frequencies that your target animals can generate and perceive. Don't underestimate the complexity of this first requirement.

Evidence suggests that homing pigeons, elephants and whales use extreme infra-sonic vibrations. You'll need equipment with a very long base-line and capture length in order to accurately capture this information. Normally recording equipment is inadequate.

Dogs and many other animals here well into the ultra-sonic which implies both special microphones, and a very fast digital sample rate for audio capture in order to adequately distinguish information in this range. Again normal human-range audio setups won't work.

We are talking orders of magnitude greater processing power and memory than you would use for normal human-range audio.

If you skimp on this step you will likely be losing key information content that will make the rest of the project impossible.

2. Record a huge sample set of animal speech. (And by this I mean thousands of hours for each species.)

3. Divide your sample data into arbitrary quarters. Take one quarter and call that your training set. The other 3 quarters are test sets you'll use to validate your training.

4. Hire Dr. Dolittle. (If he knows of anyone else who understands animal speech, then hire them too.)

5. Have Dr. Dolittle listen to every sample you've taken and tell you what the correct human translation is for that sample.

6. Build a neural network and train it using your sample data.

7. Validate your neural network against one of the test sets.

8. If this doesn't work, then examine your data and see if there is a better way to represent the data. Use Fourier transforms to represent data in the frequency domain. Look at the research in the field of human speech recognition in terms of extracting and representing charactersitics of sound. Use these characteristics as input to your neural net.

9. Iterate on 6, 7, and 8. Until you can pass the valiation test. Come up with any bright idea necessary.

10. Validate your validation using the other two sets. If that validation fails, iterate 6, 7, 8 and 9 until it passes.

11. Now you have the animal speech translated into human words -- if that's all you needed then you are done. If you want to translate it into other animal langauge, then:

12. Using your data sets generated earlier, with advice from Dr. Doolittle, break them data down into individual concepts and then use these to do a litteral translation from human to animal. It will likely be a rudimentary translation at best, but you can always improve on it by getting Dr. Dolittle to define the grammar rules for each animal language.

NOTE: If Dr. Doolittle is unavailable, and you only want to go from Pig to Dog, then you could potentially get away with hiring a Pig that speaks Dog or a Dog that speaks Pig -- however, that will probably make it difficult for you to sidecheck your progress in improving your translation and make it almost impossible to gain any intuition about your program.

Q.E.D.