Skip to content

armgit / txtai

Public Imported from GitHub
0

💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows

Add new methods for audio signal processing, closes #855

davidmezzetti39609f3ef935
diff --git a/src/python/txtai/pipeline/audio/microphone.py b/src/python/txtai/pipeline/audio/microphone.py
index f9eba6c8..d09c4882 100644
--- a/src/python/txtai/pipeline/audio/microphone.py
+++ b/src/python/txtai/pipeline/audio/microphone.py
@@ -11,7 +11,6 @@ try:
import sounddevice as sd
import webrtcvad
- from scipy.fft import rfft, rfftfreq
from scipy.signal import butter, sosfilt
from .signal import Signal, SCIPY
@@ -230,20 +229,8 @@ class Microphone(Pipeline):
True if speech is detected, False otherwise
"""
- # Calculate signal frequency
- frequency = rfftfreq(len(audio), 1.0 / self.rate)
- frequency = frequency[1:]
-
- # Calculate signal energy using amplitude
- energy = np.abs(rfft(audio))
- energy = energy[1:]
- energy = energy**2
-
- # Get energy for each frequency
- energyfreq = {}
- for x, freq in enumerate(frequency):
- if abs(freq) not in energyfreq:
- energyfreq[abs(freq)] = energy[x] * 2
+ # Calculate signal energy
+ energyfreq = Signal.energy(audio, self.rate)
# Sum speech energy
speechenergy = 0
diff --git a/src/python/txtai/pipeline/audio/signal.py b/src/python/txtai/pipeline/audio/signal.py
index 43f76ca6..dd0980f5 100644
--- a/src/python/txtai/pipeline/audio/signal.py
+++ b/src/python/txtai/pipeline/audio/signal.py
@@ -7,6 +7,7 @@ import numpy as np
# Conditional import
try:
from scipy import signal
+ from scipy.fft import rfft, rfftfreq
SCIPY = True
except ImportError:
@@ -116,3 +117,70 @@ class Signal:
# Mix audio together
return small + large
+
+ @staticmethod
+ def energy(audio, rate):
+ """
+ Calculates the signal energy for the input audio. Energy is defined as:
+
+ Energy = 2 * Signal Amplitude
+
+ Args:
+ audio: audio data
+ rate: sample rate
+
+ Returns:
+ {frequency: energy at that frequency}
+ """
+
+ # Calculate signal frequency
+ frequency = rfftfreq(len(audio), 1.0 / rate)
+ frequency = frequency[1:]
+
+ # Calculate signal energy using amplitude
+ energy = np.abs(rfft(audio))
+ energy = energy[1:]
+ energy = energy**2
+
+ # Get energy for each frequency
+ energyfreq = {}
+ for x, freq in enumerate(frequency):
+ if abs(freq) not in energyfreq:
+ energyfreq[abs(freq)] = energy[x] * 2
+
+ return energyfreq
+
+ @staticmethod
+ def trim(audio, rate, threshold=1, leading=True, trailing=True):
+ """
+ Removes leading and trailing silence from audio data.
+
+ Args:
+ audio: audio data
+ rate: sample rate
+ threshold: energy below this level will be considered silence, defaults to 1.0
+ leading: trim leading silence, defaults to True
+ trailing: trim trailing silence, defauls to True
+
+ Returns:
+ audio with silence removed
+ """
+
+ # Process in 20ms chunks
+ n, offset = int(rate * (20 / 1000.0) * 2), 0
+
+ chunks = []
+ while offset + n <= len(audio):
+ # Calculate energy for chunk and detection result
+ chunk = audio[offset : offset + n]
+ energyfreq = Signal.energy(chunk, rate)
+ chunks.append((chunk, sum(energyfreq.values()) >= threshold))
+
+ offset += n
+
+ # Find first and last active chunks
+ start = next((i for i, (_, active) in enumerate(chunks) if active), 0) if leading else 0
+ end = (len(chunks) - next((i for i, (_, active) in enumerate(chunks[::-1]) if active), 0)) if trailing else len(chunks)
+
+ # Concatenate active audio
+ return np.concatenate([chunk for chunk, _ in chunks[start:end]])