speech_recognition
speech_recognition copied to clipboard
Uberi
Custom Audio Source
I am trying to make a custom audio source class that will accept any Raw Audio. I have a stream class in my source class that can be written to by things outside of the class and then a read function for the recognizer to use. The problem is my buffer keeps running out. How do the other audio sources avoid this problem as it seems the Microphone Source has an unlimited amount of data available? Or at least whenever read is called on Microphone Source something is returned. Here is the implementation of my RawStream. Any advice would be appreciated.
class RawStream(object):
def __init__(self, sample_width=2):
self.sample_width = sample_width
self.sampleDataType = sampleDataTypes[sample_width]
self.stream = np.array([], dtype=self.sampleDataType)
# read size bytes from the beginning of stream if possible, or return None otherwise
def read(self, size):
data = self.stream[:size]
self.stream = self.stream[size:]
return data
def write(self, data):
data = np.frombuffer(data, dtype=self.sampleDataType)
self.stream = np.concatenate((self.stream, data))
Hey, I was wondering if you have found a solution to this. Currently, I'm looking for a solution to use the speech_recognition package, with my source being a ROS2 /audio topic that I would like to convert to text.
If you have any pointers for me, it would also be greatly appreciated.
The implementation of RawStream object is not thread-safe, if a write operation happens during read operation (or vice versa), the buffer get corrupted. I'd suggest just use io.BytesIO as stream rather than dealing the whole io thing yourself.
