v0.5.2

Improvements and fixes

• Disabled echo cancellation, noise suppression and auto gain control by default to match the previous library behavior.

v0.5.1

Improvements and fixes

• Fixed a crash when passing audio frames to VirtualMicrophone.write_frames() that require padding (i.e. non-multiple of 10ms worth of audio frames).

v0.5.0

New features

• Support for non-blocking virtual audio devices. This allows integration with hardware devices (e.g. via PyAudio).
• Echo cancellation, noise suppression and auto gain control can now be enabled for virtual microphones via custom constraints.
• It is now possible to pass additional Deepgram settings to start_transcription() using the new extra field.
• New demos to show how to integrate with PyAudio, how to send images and other improvements in existing demos.

Improvements and fixes

• Fixed an issue on user leave (manual or by the server) that would prevent the user to rejoin.

API changes

• Transcription defaults have been removed in favor of Deepgram's defaults. This allows to simply specify {"model": "nova-2"}.
• Transcription redact can now also be a list of strings as supported by Deepgram (e.g. ["pci"]).

v0.4.0

New features

• Added support for capturing individual participant audio tracks.
• Added support for ejecting participants.
• Support python >= 3.7 and, on Linux, glibc >= 2.28.

Improvements and fixes

• Improved general video renderer performance.
• Improved media subscriptions stability and performance.
• Fixed a deadlock that would not allow receiving multiple simultaneous video renderers.
• Fixed a deadlock when a screen share was stopped.
• Fixed an issue where setting the user name could not be reflected automatically when requesting participants list.
• Fixed an issue that could cause joins/reconnects to not complete successfully.
• Fixed a sporadic crash that could occur when handling media streams.
• Added Qt demo (similar to the existing Gtk demo).
• Qt and Gtk demos can now save the selected participant audio into a WAV file and can also render screen share.

v0.3.1

Improvements and fixes

• Fixed an issue that could cause daily-python clients to join a meeting session in a different region.
• Fixed a dead-lock that could occur when a CallClient is destroyed.

v0.3.0

New features

• Support for sending chat messages to Daily Prebuilt (CallClient.send_prebuilt_chat_message()).
• Added Python type hints (helpful for editor completions).
• Support for Python 3.8.

Improvements and fixes

• Improved response time of CallClient getter functions.
• Improved low-latency performace of virtual audio devices.
• Fixed potential crash after CallClient.leave().
• Improved internal safeness of participant video renderers.
• Fixed a VirtualMicrophoneDevice memory leak.
• Properly trigger a transcription error event if transcription can't start.
• Demos have been updated to show more real live code.

API changes

• Removed detect_language from TranscriptionSettings.
• EventHandler.on_transcription_stopped can now tell if transcription was stopped by a user or because of an error occurred.

v0.2.0

New features

• Support for start/stop recordings.
• Support for start/stop transcriptions and receive transcriptions messages.

Improvements and fixes

• Fixed an issue that was causing sporadic audio gaps on macOS and in certain OS task scheduling scenarios.
• Network re-connections have been improved.

API changes

• VirtualSpeakerDevice.read_frames() has been improved and doesn't require the user to add sleeps. Therefore, it is now possible to read, for example, 10 seconds of audio in a single call. Since the timings are now controlled internally, this minimizes any potential audio issues.

The following old code:

SAMPLE_RATE = 16000
READ_INTERVAL = 0.01
FRAMES_TO_READ = int(SAMPLE_RATE * READ_INTERVAL)
SECONDS_TO_READ = 10.0

for _ in range (int(SECONDS_TO_READ / READ_INTERVAL)):
  buffer = speaker.read_frames(FRAMES_TO_READ)
  time.sleep(READ_INTERVAL)

can be replaced with:

SECONDS_TO_READ = 10
FRAMES_TO_READ = SAMPLE_RATE * SECONDS_TO_READ
buffer = speaker.read_frames(FRAMES_TO_READ)

v0.1.1

• Fixed an issue where virtual devices could cause other Python threads to be blocked.

v0.1.0

• Fixed an issue where meeting_token was not properly handled in CallClient.join().

• VirtualMicrophoneDevice.write_samples() has been renamed to VirtualMicrophoneDevice.write_frames()

• VirtualMicrophoneDevice.read_samples() has been renamed to VirtualMicrophoneDevice.read_frames()

• VirtualMicrophoneDevice.write_frames() now takes a single argument frames. frames are audio frames, an interleaved set of audio samples per channel. write_frames() is a synchronous call that will finish when all audio frames have been written. write_frames() also adds padding when necessary.

• Handling microphone audio samples has been simplified. Previous, the following code was required:

while True:
  buffer = stream.read(960)

  if not buffer:
    break

  if len(buffer) < 960:
    buffer += b'\00' * (960 - len(buffer))

  microphone.write_samples(buffer, 480)

  time.sleep(0.03)

It can now be replaced with:

microphone.write_frames(stream.read())

v0.0.15

• Virtual camera support. It is now possible to create a virtual camera and dynamically provide frames to it so they are sent as your client camera to the meeting.
• Color formats have been renamed ARGB32 -> ARGB, BGRA32 -> BGRA........ RGB24-> RGB. The bytestrings with frame contents are now expected to follow the color format order. So, for example, an RGBA image will have R in the first byte, G in the second byte, B in the third byte, A in the fourth byte and so on.
• Daily.init() parameter virtual_devices is not necessary anymore. Therefore, Daily.init(virtual_devices = True) is now simply Daily.init().