Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Align start seconds for transcript segment #906

Merged
merged 10 commits into from
Sep 22, 2024
Merged

Align start seconds for transcript segment #906

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
2229df9
Align start seconds for transcript segment
beastoin Sep 22, 2024
60376d9
Trim the audio frames regarding the start, end
beastoin Sep 22, 2024
8010780
Create memory with delta segment start
beastoin Sep 22, 2024
a306ce5
Remove redundant len check on new segments
beastoin Sep 22, 2024
978957d
Use transcript timer segment start instead of timer start(audio) for …
beastoin Sep 22, 2024
bd95410
Backward compatbile with the old ProcessingMemory model
beastoin Sep 22, 2024
8465de6
Trim audio frames
beastoin Sep 22, 2024
60b33c4
Use timmer start to validate create new memory
beastoin Sep 22, 2024
35027e2
Update finished at for combined memory
beastoin Sep 22, 2024
461d062
Add silent seconds between 2 combining audio
beastoin Sep 22, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions backend/database/memories.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,6 +138,11 @@ def update_memory_events(uid: str, memory_id: str, events: List[dict]):
memory_ref = user_ref.collection('memories').document(memory_id)
memory_ref.update({'structured.events': events})

def update_memory_finished_at(uid: str, memory_id: str, finished_at: datetime):
user_ref = db.collection('users').document(uid)
memory_ref = user_ref.collection('memories').document(memory_id)
memory_ref.update({'finished_at': finished_at})


# VISBILITY

Expand Down
1 change: 1 addition & 0 deletions backend/models/processing_memory.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@ class ProcessingMemory(BaseModel):
audio_url: Optional[str] = None
created_at: datetime
timer_start: float
timer_segment_start: Optional[float] = None
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The addition of timer_segment_start as an optional field in the ProcessingMemory class seems fine. However, it would be beneficial to add a comment explaining what this new field represents and how it differs from timer_start. This will help other developers understand its purpose more easily.

-    timer_segment_start: Optional[float] = None
+    timer_segment_start: Optional[float] = None  # Start time for a specific transcript segment

timer_starts: List[float] = []
language: Optional[str] = None # applies only to Friend # TODO: once released migrate db to default 'en'
transcript_segments: List[TranscriptSegment] = []
Expand Down
61 changes: 48 additions & 13 deletions backend/routers/transcribe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
import threading
import math
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The math module is imported but not used in the provided code. Please ensure that it's used in the complete file or remove the import to keep the code clean.

- import math

import asyncio
import time
from typing import List
Expand Down Expand Up @@ -139,10 +140,27 @@ def stream_transcript(segments, stream_id):
nonlocal processing_memory
nonlocal processing_memory_synced
nonlocal memory_transcript_segements
nonlocal segment_start
nonlocal segment_end

if not segments or len(segments) == 0:
return

# Align the start, end segment
if not segment_start:
start = segments[0]["start"]
segment_start = start

# end
end = segments[-1]["end"]
if not segment_end or segment_end < end:
segment_end = end

for i, segment in enumerate(segments):
segment["start"] -= segment_start
segment["end"] -= segment_start
segments[i] = segment

Comment on lines +143 to +163
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The logic for aligning the start and end of segments seems correct. However, there is no error handling or validation for the segments data. It would be good to add some checks to ensure that each segment has a "start" and "end" field before trying to access them. This will prevent potential KeyError exceptions.

        # Align the start, end segment
        if not segment_start:
            start = segments[0].get("start")
            if start is None:
                raise ValueError("Each segment should have a 'start' field.")
            segment_start = start

        # end
        end = segments[-1].get("end")
        if end is None:
            raise ValueError("Each segment should have an 'end' field.")
        if not segment_end or segment_end < end:
            segment_end = end

        for i, segment in enumerate(segments):
            if "start" not in segment or "end" not in segment:
                raise ValueError("Each segment should have 'start' and 'end' fields.")
            segment["start"] -= segment_start
            segment["end"] -= segment_start
            segments[i] = segment

asyncio.run_coroutine_threadsafe(websocket.send_json(segments), loop)
threading.Thread(target=process_segments, args=(uid, segments)).start()

Expand Down Expand Up @@ -178,6 +196,9 @@ def stream_audio(audio_buffer):
websocket_active = True
websocket_close_code = 1001 # Going Away, don't close with good from backend
timer_start = None
segment_start = None
segment_end = None
audio_frames_per_sec = 100
# audio_buffer = None
duration = 0
try:
Expand Down Expand Up @@ -239,7 +260,8 @@ async def receive_audio(dg_socket1, dg_socket2, soniox_socket, speechmatics_sock
# audio_file = open(path, "a")
try:
while websocket_active:
data = await websocket.receive_bytes()
raw_data = await websocket.receive_bytes()
data = raw_data[:]
Comment on lines +263 to +264
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

It seems like you're creating a copy of raw_data with data = raw_data[:]. If raw_data is large, this could be a performance issue. If you're not modifying raw_data later, you can use it directly instead of creating a copy.

- data = raw_data[:]

# audio_buffer.extend(data)
if codec == 'opus' and sample_rate == 16000:
data = decoder.decode(bytes(data), frame_size=160)
Expand All @@ -262,7 +284,7 @@ async def receive_audio(dg_socket1, dg_socket2, soniox_socket, speechmatics_sock
dg_socket2.send(data)

# stream
stream_audio(data)
stream_audio(raw_data)
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The function stream_audio is now being called with raw_data instead of data. Ensure that this change doesn't affect the functionality of stream_audio, as it was previously receiving the potentially decoded data.

- stream_audio(data)
+ stream_audio(raw_data)


# audio_buffer = bytearray()

Expand Down Expand Up @@ -331,10 +353,11 @@ async def _create_processing_memory():
last_processing_memory_data = processing_memories_db.get_last(uid)
if last_processing_memory_data:
last_processing_memory = ProcessingMemory(**last_processing_memory_data)
segment_end = 0
last_segment_end = 0
for segment in last_processing_memory.transcript_segments:
segment_end = max(segment_end, segment.end)
if last_processing_memory.timer_start + segment_end + min_seconds_limit > time.time():
last_segment_end = max(last_segment_end, segment.end)
timer_segment_start = last_processing_memory.timer_segment_start if last_processing_memory.timer_segment_start else last_processing_memory.timer_start
if timer_segment_start + last_segment_end + min_seconds_limit > time.time():
Comment on lines +356 to +360
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The variable name last_segment_end is misleading because it suggests that it's the end time of the last segment, but it's actually the maximum end time across all segments. Consider renaming it to something like max_segment_end for clarity.

- last_segment_end = 0
+ max_segment_end = 0
- for segment in last_processing_memory.transcript_segments:
-     last_segment_end = max(last_segment_end, segment.end)
+ for segment in last_processing_memory.transcript_segments:
+     max_segment_end = max(max_segment_end, segment.end)

processing_memory = last_processing_memory

# Or create new
Expand All @@ -343,6 +366,7 @@ async def _create_processing_memory():
id=str(uuid.uuid4()),
created_at=datetime.now(timezone.utc),
timer_start=timer_start,
timer_segment_start=timer_start+segment_start,
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

This line assumes that segment_start is not None, but there doesn't seem to be a guarantee of this in the code. If segment_start is None, this will raise a TypeError. Please add a check for None before this operation.

language=language,
)

Expand Down Expand Up @@ -375,11 +399,22 @@ async def _post_process_memory(memory: Memory):
nonlocal processing_memory
nonlocal processing_audio_frames
nonlocal processing_audio_frame_synced
nonlocal segment_start
nonlocal segment_end

# Create wav
processing_audio_frame_synced = len(processing_audio_frames)

# Remove audio frames [start, end]
left = 0
if segment_start:
left = max(left, math.floor(segment_start) * audio_frames_per_sec)
right = processing_audio_frame_synced
if segment_end:
right = min(math.ceil(segment_end) * audio_frames_per_sec, right)

Comment on lines +402 to +415
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The logic for creating a WAV file seems correct. However, there is no error handling or validation for the segment_start and segment_end variables. If these variables are not numbers, this will raise a TypeError. Consider adding error handling or validation to ensure the expected types.

        if segment_start:
+            if not isinstance(segment_start, (int, float)):
+                raise TypeError("`segment_start` should be a number.")
            left = max(left, math.floor(segment_start) * audio_frames_per_sec)
        if segment_end:
+            if not isinstance(segment_end, (int, float)):
+                raise TypeError("`segment_end` should be a number.")
            right = min(math.ceil(segment_end) * audio_frames_per_sec, right)

file_path = f"_temp/{memory.id}_{uuid.uuid4()}_be"
create_wav_from_bytes(file_path=file_path, frames=processing_audio_frames[:processing_audio_frame_synced],
create_wav_from_bytes(file_path=file_path, frames=processing_audio_frames[left:right],
Comment on lines +402 to +417
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The logic for creating a WAV file from a subset of audio frames based on segment start and end times seems correct. However, there is no error handling or validation for the segment_start and segment_end variables. It would be good to add some checks to ensure that they are within valid bounds before using them to slice the processing_audio_frames list. This will prevent potential IndexError exceptions.

        # Remove audio frames [start, end]
        left = 0
        if segment_start:
            left = max(left, math.floor(segment_start) * audio_frames_per_sec)
        right = processing_audio_frame_synced
        if segment_end:
            right = min(math.ceil(segment_end) * audio_frames_per_sec, right)

         file_path = f"_temp/{memory.id}_{uuid.uuid4()}_be"
        create_wav_from_bytes(file_path=file_path, frames=processing_audio_frames[left:right],
                               frame_rate=sample_rate, channels=channels, codec=codec, )

frame_rate=sample_rate, channels=channels, codec=codec, )

# Try merge new audio with the previous
Expand Down Expand Up @@ -489,6 +524,10 @@ async def _create_memory():
memories_db.update_memory_segments(uid, memory.id,
[segment.dict() for segment in memory.transcript_segments])

# Update finished at
memory.finished_at = datetime.fromtimestamp(memory.started_at.timestamp() + processing_memory.transcript_segments[-1].end, timezone.utc)
memories_db.update_memory_finished_at(uid, memory.id, memory.finished_at)

# Process
memory = process_memory(uid, memory.language, memory, force_process=True)

Expand Down Expand Up @@ -525,6 +564,8 @@ async def _try_flush_new_memory_with_lock(time_validate: bool = True):
async def _try_flush_new_memory(time_validate: bool = True):
nonlocal memory_transcript_segements
nonlocal timer_start
nonlocal segment_start
nonlocal segment_end
nonlocal processing_memory
nonlocal processing_memory_synced
nonlocal processing_audio_frames
Expand All @@ -535,13 +576,8 @@ async def _try_flush_new_memory(time_validate: bool = True):
return

# Validate last segment
last_segment = None
if len(memory_transcript_segements) > 0:
last_segment = memory_transcript_segements[-1]
if not last_segment:
if not segment_end:
print("Not last segment or last segment invalid")
Comment on lines +580 to 581
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The validation for segment_end seems to be incomplete. It checks if segment_end is not None but does not handle the case when it is None. Consider adding appropriate error handling or logging for this case.

        if not segment_end:
             print("Not last segment or last segment invalid")
+             return

if last_segment:
print(f"{last_segment.dict()}")
return

# First chunk, create processing memory
Expand All @@ -552,7 +588,6 @@ async def _try_flush_new_memory(time_validate: bool = True):

# Validate transcript
# Longer 120s
segment_end = last_segment.end
now = time.time()
should_create_memory_time = True
if time_validate:
Comment on lines 589 to 594
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The variable now is defined but not used in the provided code. Please ensure that it's used in the complete file or remove the declaration to keep the code clean.

- now = time.time()

Expand Down
6 changes: 3 additions & 3 deletions backend/utils/processing_memories.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,11 +23,11 @@ async def create_memory_by_processing_memory(uid: str, processing_memory_id: str
if not transcript_segments or len(transcript_segments) == 0:
print("Transcript segments is invalid")
return
timer_start = processing_memory.timer_start
timer_segment_start = processing_memory.timer_segment_start if processing_memory.timer_segment_start else processing_memory.timer_start
segment_end = transcript_segments[-1].end
new_memory = CreateMemory(
started_at=datetime.fromtimestamp(timer_start, timezone.utc),
finished_at=datetime.fromtimestamp(timer_start + segment_end, timezone.utc),
started_at=datetime.fromtimestamp(timer_segment_start, timezone.utc),
finished_at=datetime.fromtimestamp(timer_segment_start + segment_end, timezone.utc),
Comment on lines +26 to +30
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Image description Entelligence.AI

The changes made here are logical and seem to be in line with the PR's goal of aligning start seconds for transcript segments. However, there is a potential issue with error handling. If processing_memory.timer_segment_start or processing_memory.timer_start is None, this code will throw an exception when trying to create a timestamp from None. It would be better to handle this case explicitly.

- timer_segment_start = processing_memory.timer_segment_start if processing_memory.timer_segment_start else processing_memory.timer_start
+ timer_segment_start = processing_memory.timer_segment_start if processing_memory.timer_segment_start else processing_memory.timer_start
+ if timer_segment_start is None:
+     print("Timer segment start is invalid")
+     return

This way, we can avoid unexpected exceptions and provide a more informative error message.

language=processing_memory.language,
transcript_segments=transcript_segments,
)
Expand Down
Loading