{-# LANGUAGE Arrows #-} {-# LANGUAGE DataKinds #-} {-# LANGUAGE FlexibleInstances #-} {-# LANGUAGE KindSignatures #-} {-# LANGUAGE MultiParamTypeClasses #-} {-# LANGUAGE TypeFamilies #-} -- {-# OPTIONS_GHC -Wno-unticked-promoted-constructors #-} -- TODO Find out exact version of cabal? GHC? that have a problem with this module FRP.Rhine.Clock.Realtime.Audio ( AudioClock (..) , AudioRate (..) , PureAudioClock (..) , pureAudioClockF ) where -- base import GHC.Float (double2Float) import GHC.TypeLits (Nat, natVal, KnownNat) import Data.Time.Clock -- transformers import Control.Monad.IO.Class -- dunai import Control.Monad.Trans.MSF.Except -- rhine import FRP.Rhine -- | Rates at which audio signals are typically sampled. data AudioRate = Hz44100 | Hz48000 | Hz96000 -- | Converts an 'AudioRate' to its corresponding rate as an 'Integral'. rateToIntegral :: Integral a => AudioRate -> a rateToIntegral Hz44100 = 44100 rateToIntegral Hz48000 = 48000 rateToIntegral Hz96000 = 96000 -- TODO Test extensively {- | A clock for audio analysis and synthesis. It internally processes samples in buffers of size 'bufferSize', (the programmer does not have to worry about this), at a sample rate of 'rate' (of type 'AudioRate'). Both these parameters are in the type signature, so it is not possible to compose signals with different buffer sizes or sample rates. After processing a buffer, the clock will wait the remaining time until the next buffer must be processed, using system UTC time. The tag of the clock specifies whether the attempt to finish the last buffer in real time was successful. A value of 'Nothing' represents success, a value of @Just double@ represents a lag of 'double' seconds. -} data AudioClock (rate :: AudioRate) (bufferSize :: Nat) = AudioClock class AudioClockRate (rate :: AudioRate) where theRate :: AudioClock rate bufferSize -> AudioRate theRateIntegral :: Integral a => AudioClock rate bufferSize -> a theRateIntegral = rateToIntegral . theRate theRateNum :: Num a => AudioClock rate bufferSize -> a theRateNum = fromInteger . theRateIntegral instance AudioClockRate Hz44100 where theRate _ = Hz44100 instance AudioClockRate Hz48000 where theRate _ = Hz48000 instance AudioClockRate Hz96000 where theRate _ = Hz96000 theBufferSize :: (KnownNat bufferSize, Integral a) => AudioClock rate bufferSize -> a theBufferSize = fromInteger . natVal instance (MonadIO m, KnownNat bufferSize, AudioClockRate rate) => Clock m (AudioClock rate bufferSize) where type TimeDomainOf (AudioClock rate bufferSize) = UTCTime type Tag (AudioClock rate bufferSize) = Maybe Double startClock audioClock = do let step = picosecondsToDiffTime -- The only sufficiently precise conversion function $ round (10 ^ (12 :: Integer) / theRateNum audioClock :: Double) bufferSize = theBufferSize audioClock runningClock :: MonadIO m => UTCTime -> Maybe Double -> MSF m () (UTCTime, Maybe Double) runningClock initialTime maybeWasLate = safely $ do bufferFullTime <- try $ proc () -> do n <- count -< () let nextTime = (realToFrac step * fromIntegral (n :: Int)) `addUTCTime` initialTime _ <- throwOn' -< (n >= bufferSize, nextTime) returnA -< (nextTime, if n == 0 then maybeWasLate else Nothing) currentTime <- once_ $ liftIO getCurrentTime let lateDiff = currentTime `diffTime` bufferFullTime late = if lateDiff > 0 then Just lateDiff else Nothing safe $ runningClock bufferFullTime late initialTime <- liftIO getCurrentTime return ( runningClock initialTime Nothing , initialTime ) {- | A side-effect free clock for audio synthesis and analysis. The sample rate is given by 'rate' (of type 'AudioRate'). Since this clock does not wait for the completion of buffers, the producer or the consumer of the signal has the obligation to synchronise the signal with the system clock, if realtime is desired. Otherwise, the clock is also suitable e.g. for batch processing of audio files. -} data PureAudioClock (rate :: AudioRate) = PureAudioClock class PureAudioClockRate (rate :: AudioRate) where thePureRate :: PureAudioClock rate -> AudioRate thePureRateIntegral :: Integral a => PureAudioClock rate -> a thePureRateIntegral = rateToIntegral . thePureRate thePureRateNum :: Num a => PureAudioClock rate -> a thePureRateNum = fromInteger . thePureRateIntegral instance (Monad m, PureAudioClockRate rate) => Clock m (PureAudioClock rate) where type TimeDomainOf (PureAudioClock rate) = Double type Tag (PureAudioClock rate) = () startClock audioClock = return ( arr (const (1 / thePureRateNum audioClock)) >>> sumS &&& arr (const ()) , 0 ) -- | A rescaled version of 'PureAudioClock' with 'TimeDomain' 'Float'. type PureAudioClockF (rate :: AudioRate) = RescaledClock (PureAudioClock rate) Float pureAudioClockF :: PureAudioClockF rate pureAudioClockF = RescaledClock { unscaledClock = PureAudioClock , rescale = double2Float }