def process_with_initialised_plugin(ff, sample_rate, step_size, plugin, outputs): out_indices = dict([(id, plugin.get_output(id)["output_index"]) for id in outputs]) plugin.reset() fi = 0 for f in ff: timestamp = vampyhost.frame_to_realtime(fi, sample_rate) results = plugin.process_block(f, timestamp) # results is a dict mapping output number -> list of feature dicts for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield {o: r} fi = fi + step_size results = plugin.get_remaining_features() for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield {o: r}
def get_feature_step_time(sample_rate, step_size, output_desc): if output_desc["sampleType"] == vampyhost.ONE_SAMPLE_PER_STEP: return vampyhost.frame_to_realtime(step_size, sample_rate) elif output_desc["sampleType"] == vampyhost.FIXED_SAMPLE_RATE: return vampyhost.RealTime('seconds', 1.0 / output_desc["sampleRate"]) else: return 1
def process(self, frames, eod=False): timestamp = vampyhost.frame_to_realtime(self.frame_index, self.input_samplerate) results = self.plugin.process_block(frames.T, timestamp) if self.out_index in results: for res in results[self.out_index]: self.vamp_results.append({self.plugin_output: res}) self.frame_index += self.input_stepsize return frames, eod
def timestamp_features(sample_rate, step_size, output_desc, features): n = -1 if output_desc["sampleType"] == vampyhost.ONE_SAMPLE_PER_STEP: for f in features: n = n + 1 t = vampyhost.frame_to_realtime(n * step_size, sample_rate) f["timestamp"] = t yield f elif output_desc["sampleType"] == vampyhost.FIXED_SAMPLE_RATE: output_rate = output_desc["sampleRate"] for f in features: if "has_timestamp" in f: n = int(f["timestamp"].to_float() * output_rate + 0.5) else: n = n + 1 f["timestamp"] = vampyhost.RealTime('seconds', float(n) / output_rate) yield f else: for f in features: yield f
def process_with_initialised_plugin(ff, sample_rate, step_size, plugin, outputs): out_indices = dict([(id, plugin.get_output(id)["output_index"]) for id in outputs]) plugin.reset() fi = 0 for f in ff: timestamp = vampyhost.frame_to_realtime(fi, sample_rate) results = plugin.process_block(f, timestamp) # results is a dict mapping output number -> list of feature dicts for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield { o: r } fi = fi + step_size results = plugin.get_remaining_features() for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield { o: r }
def process_frames_multiple_outputs(ff, sample_rate, step_size, plugin_key, outputs, parameters={}): """Process audio data with a Vamp plugin, and make the results from a set of plugin outputs available as a generator. The provided data should be an enumerable sequence of time-domain audio frames, of which each frame is 2-dimensional list or NumPy array of floats. The first dimension is taken to be the channel count, and the second dimension the frame or block size. The step_size argument gives the increment in audio samples from one frame to the next. Each frame must have the same size. The returned results will be those calculated by the plugin with the given key and returned through its outputs whose identifiers are given in the outputs argument. If the parameters dict is non-empty, the plugin will be configured by setting its parameters according to the (string) key and (float) value data found in the dict. This function acts as a generator, yielding a sequence of result feature sets as it obtains them. Each feature set is a dictionary mapping from output identifier to a list of features, each represented as a dictionary containing, optionally, timestamp and duration (RealTime objects), label (string), and a 1-dimensional array of float values. """ plugin = vampyhost.load_plugin( plugin_key, sample_rate, vampyhost.ADAPT_INPUT_DOMAIN + vampyhost.ADAPT_BUFFER_SIZE + vampyhost.ADAPT_CHANNEL_COUNT) out_indices = dict([(id, plugin.get_output(id)["output_index"]) for id in outputs]) fi = 0 channels = 0 block_size = 0 for f in ff: if fi == 0: channels = f.shape[0] block_size = f.shape[1] plugin.set_parameter_values(parameters) if not plugin.initialise(channels, step_size, block_size): raise "Failed to initialise plugin" timestamp = vampyhost.frame_to_realtime(fi, sample_rate) results = plugin.process_block(f, timestamp) # results is a dict mapping output number -> list of feature dicts for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield {o: r} fi = fi + step_size if fi > 0: results = plugin.get_remaining_features() for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield {o: r} plugin.unload()
def process_frames_multiple_outputs(ff, sample_rate, step_size, plugin_key, outputs, parameters = {}): """Process audio data with a Vamp plugin, and make the results from a set of plugin outputs available as a generator. The provided data should be an enumerable sequence of time-domain audio frames, of which each frame is 2-dimensional list or NumPy array of floats. The first dimension is taken to be the channel count, and the second dimension the frame or block size. The step_size argument gives the increment in audio samples from one frame to the next. Each frame must have the same size. The returned results will be those calculated by the plugin with the given key and returned through its outputs whose identifiers are given in the outputs argument. If the parameters dict is non-empty, the plugin will be configured by setting its parameters according to the (string) key and (float) value data found in the dict. This function acts as a generator, yielding a sequence of result feature sets as it obtains them. Each feature set is a dictionary mapping from output identifier to a list of features, each represented as a dictionary containing, optionally, timestamp and duration (RealTime objects), label (string), and a 1-dimensional array of float values. """ plugin = vampyhost.load_plugin(plugin_key, sample_rate, vampyhost.ADAPT_INPUT_DOMAIN + vampyhost.ADAPT_BUFFER_SIZE + vampyhost.ADAPT_CHANNEL_COUNT) out_indices = dict([(id, plugin.get_output(id)["output_index"]) for id in outputs]) fi = 0 channels = 0 block_size = 0 for f in ff: if fi == 0: channels = f.shape[0] block_size = f.shape[1] plugin.set_parameter_values(parameters) if not plugin.initialise(channels, step_size, block_size): raise "Failed to initialise plugin" timestamp = vampyhost.frame_to_realtime(fi, sample_rate) results = plugin.process_block(f, timestamp) # results is a dict mapping output number -> list of feature dicts for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield { o: r } fi = fi + step_size if fi > 0: results = plugin.get_remaining_features() for o in outputs: ix = out_indices[o] if ix in results: for r in results[ix]: yield { o: r } plugin.unload()