Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
19 changes: 19 additions & 0 deletions examples/test-upliftai.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,19 @@
"""Example usage of the UpliftAI engine."""

import os

from tts_wrapper import UpliftAIClient


def main() -> None:
api_key = os.getenv("UPLIFTAI_KEY")
if not api_key:
raise RuntimeError("UPLIFTAI_KEY environment variable is not set")

client = UpliftAIClient(api_key=api_key)
text = "Testing the UpliftAI text to speech engine"
client.speak_streamed(text)


if __name__ == "__main__":
main()
1 change: 1 addition & 0 deletions tests/load_credentials.py
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
"elevenlabs": ["ELEVENLABS_API_KEY"],
"witai": ["WITAI_TOKEN"],
"playht": ["PLAYHT_API_KEY", "PLAYHT_USER_ID"],
"upliftai": ["UPLIFTAI_KEY"],
}


Expand Down
5 changes: 5 additions & 0 deletions tests/test_generate_audiofile.py
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
GoogleClient,
GoogleTransClient,
MicrosoftClient,
UpliftAIClient,
PollyClient,
SherpaOnnxClient,
WatsonClient,
Expand Down Expand Up @@ -48,6 +49,10 @@
"client": WitAiClient,
"credential_keys": ["WITAI_TOKEN"],
},
"upliftai": {
"client": UpliftAIClient,
"credential_keys": ["UPLIFTAI_KEY"],
},
"googletrans": {
"client_lambda": lambda: GoogleTransClient("en-co.uk"),
},
Expand Down
21 changes: 21 additions & 0 deletions tests/test_tts_engines.py
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,7 @@
PlayHTClient,
PollyClient,
SherpaOnnxClient,
UpliftAIClient,
WatsonClient,
WitAiClient,
eSpeakClient,
Expand All @@ -37,6 +38,7 @@
"espeak": eSpeakClient,
"playht": PlayHTClient,
"openai": OpenAIClient,
"upliftai": UpliftAIClient,
}

# Add AVSynth only on macOS
Expand Down Expand Up @@ -126,6 +128,8 @@ def check_credentials(service):
f"ElevenLabs API key: {elevenlabs_api_key[:5]}...{elevenlabs_api_key[-5:] if elevenlabs_api_key else ''}"
)
client = ElevenLabsClient(credentials=elevenlabs_api_key)
elif service == "upliftai":

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 | Confidence: High

The check_credentials function is already flagged by Ruff as overly complex (C901). Adding another conditional branch for UpliftAI exacerbates this maintainability issue. The function now has 25+ branches, making it difficult to modify and test. This pattern also doesn't scale well for new engines. The related context shows this function is critical for test setup across all engines.

Code Suggestion:

_CREDENTIAL_CHECKERS = {
      "upliftai": lambda: UpliftAIClient(api_key=os.getenv("UPLIFTAI_KEY")),
      # ... other services ...
  }

  def check_credentials(service):
      if service not in _CREDENTIAL_CHECKERS:
          return False
      try:
          client = _CREDENTIAL_CHECKERS[service]()
          return client.check_credentials()
      except Exception:
          return False

client = UpliftAIClient(api_key=os.getenv("UPLIFTAI_KEY"))
elif service == "witai":
client = WitAiClient(credentials=os.getenv("WITAI_API_KEY"))
elif service == "googletrans":
Expand Down Expand Up @@ -199,6 +203,8 @@ def create_tts_client(service):
return PlayHTClient(
credentials=(os.getenv("PLAYHT_API_KEY"), os.getenv("PLAYHT_USER_ID"))
)
if service == "upliftai":
return UpliftAIClient(api_key=os.getenv("UPLIFTAI_KEY"))
if service == "avsynth" and sys.platform == "darwin":
return AVSynthClient()
if service == "openai":
Expand Down Expand Up @@ -232,6 +238,8 @@ def test_synth_to_bytes(service):
audio_bytes = client.synth_to_bytes(text)
assert isinstance(audio_bytes, bytes)
assert len(audio_bytes) > 0
if service == "upliftai":
assert not audio_bytes.startswith(b"RIFF")
except Exception as e:
pytest.fail(f"Synthesis failed with error: {e}")

Expand All @@ -247,13 +255,26 @@ def test_synth_to_bytes(service):
audio_bytes = client.synth_to_bytes(ssml_text)
assert isinstance(audio_bytes, bytes)
assert len(audio_bytes) > 0
if service == "upliftai":
assert not audio_bytes.startswith(b"RIFF")
except (AttributeError, NotImplementedError):
# Skip SSML test for engines that don't support it
pass
except Exception as e:
pytest.fail(f"SSML synthesis failed with error: {e}")


@pytest.mark.synthetic

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P1 | Confidence: High

This test validates the WAV header stripping but doesn't verify the audio content remains valid after stripping. The related context shows similar tests for other engines focus on basic audio properties (byte length, format), but don't validate audio integrity. This could allow broken audio to pass tests if the header stripping corrupts the data.

Code Suggestion:

# Add to test after header check
import wave
from io import BytesIO
# ... after first_chunk assertion ...
full_audio = b''.join(stream)
try:
    with BytesIO(full_audio) as bio:
        with wave.open(bio) as wav:
        assert wav.getnchannels() == 1  # Validate basic audio properties
        assert wav.getframerate() == 22050
except Exception as e:
    pytest.fail(f"Stripped audio is invalid: {e}")

def test_upliftai_streaming_strips_wav_header():
service = "upliftai"
if not check_credentials(service):
pytest.skip("UpliftAI TTS credentials are invalid or unavailable")
client = create_tts_client(service)
stream = client.synth_to_bytestream("Streaming header test")
first_chunk = next(stream)
assert not first_chunk.startswith(b"RIFF")


@pytest.mark.synthetic
@pytest.mark.parametrize("service", TTS_CLIENTS.keys())
def test_playback_with_callbacks(service):
Expand Down
1 change: 1 addition & 0 deletions tts_wrapper/engines/__init__.py
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,7 @@
from .googletrans import *
from .microsoft import *
from .openai import *
from .upliftai import *
from .pico import *
from .playht import *
from .polly import *
Expand Down
5 changes: 5 additions & 0 deletions tts_wrapper/engines/upliftai/__init__.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
"""UpliftAI TTS engine for tts-wrapper."""

from .client import UpliftAIClient

__all__ = ["UpliftAIClient"]
183 changes: 183 additions & 0 deletions tts_wrapper/engines/upliftai/client.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,183 @@
from __future__ import annotations

import logging
import os
from pathlib import Path
from typing import TYPE_CHECKING, Any, Callable

import requests

from tts_wrapper.tts import AbstractTTS

if TYPE_CHECKING:
from collections.abc import Generator


logger = logging.getLogger(__name__)
RIFF_HEADER = b"RIFF"


class UpliftAIClient(AbstractTTS):
"""Client for the UpliftAI text-to-speech API."""

BASE_URL = "https://api.upliftai.org/v1/synthesis/text-to-speech"
STREAM_URL = f"{BASE_URL}/stream"
DEFAULT_VOICE = "v_8eelc901" # Info/Education Urdu

def __init__(self, api_key: str | None = None) -> None:
super().__init__()
self.api_key = api_key or os.getenv("UPLIFTAI_KEY")
if not self.api_key:
msg = "UpliftAI API key is required. Set UPLIFTAI_KEY or pass api_key."
raise ValueError(msg)

self.headers = {"Authorization": self.api_key, "Content-Type": "application/json"}
self.audio_rate = 22050
self.voice_id = self.DEFAULT_VOICE

def set_voice(self, voice_id: str, lang: str | None = None) -> None:
"""Set the voice for synthesis."""
self.voice_id = voice_id
if lang:
self.lang = lang

def synth_to_bytes(self, text: Any, voice_id: str | None = None) -> bytes:
"""Synthesize text to audio bytes using the non-streaming endpoint."""
voice = voice_id or self.voice_id or self.DEFAULT_VOICE
payload = {
"voiceId": voice,
"text": str(text),
"outputFormat": "WAV_22050_16",
}
response = requests.post(self.BASE_URL, json=payload, headers=self.headers, timeout=30)
response.raise_for_status()
audio_bytes = response.content
if audio_bytes[:4] == RIFF_HEADER:
audio_bytes = self._strip_wav_header(audio_bytes)
return audio_bytes

def synth_to_bytestream(

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P1 | Confidence: High

The WAV header stripping logic in the streaming method assumes a fixed 44-byte header. This is fragile because WAV headers can vary in size depending on format details. The related test (test_upliftai_streaming_strips_wav_header) only checks for the absence of "RIFF" but doesn't validate the actual audio format. If the API changes its header structure, this could break audio decoding or cause silent corruption. The non-streaming method uses a more robust RIFF signature check.

Suggested change
def synth_to_bytestream(
def synth_to_bytestream(...):
# ... same setup ...
with requests.post(...) as response:
response.raise_for_status()
# Find RIFF header dynamically
header_buffer = b""
for chunk in response.iter_content(chunk_size=4096):
if not chunk:
continue
if not header_skipped:
header_buffer += chunk
riff_pos = header_buffer.find(RIFF_HEADER)
if riff_pos == -1:
# Haven't found header yet, keep buffering
continue
# Calculate header end (RIFF + 4 bytes size + 4 bytes WAVE = 8 bytes minimum)
# Actual header size is at riff_pos+4 (little-endian uint32)
if len(header_buffer) < riff_pos + 8:
continue
header_size = int.from_bytes(header_buffer[riff_pos+4:riff_pos+8], 'little') + 8
if len(header_buffer) < riff_pos + header_size:
continue
# Yield data after header
data = header_buffer[riff_pos + header_size:]
header_skipped = True
if data:
yield data
continue
yield chunk

self, text: Any, voice_id: str | None = None
) -> Generator[bytes, None, None]:
"""Stream synthesized audio chunks from the API."""
voice = voice_id or self.voice_id or self.DEFAULT_VOICE
payload = {
"voiceId": voice,
"text": str(text),
"outputFormat": "WAV_22050_16",
}
with requests.post(
self.STREAM_URL, json=payload, headers=self.headers, stream=True, timeout=30
) as response:
response.raise_for_status()
header_buffer = b""
header_skipped = False
for chunk in response.iter_content(chunk_size=4096):
if not chunk:
continue
if not header_skipped:
header_buffer += chunk
if len(header_buffer) <= 44:
continue
data = header_buffer[44:]
header_skipped = True
if not data:
continue
yield data
continue
yield chunk

def synth(
self,
text: Any,
output_file: str | Path,
output_format: str = "wav",
voice_id: str | None = None,
) -> None:
"""Synthesize text to a file."""
audio_bytes = self.synth_to_bytes(text, voice_id)
with Path(output_file).open("wb") as f:
f.write(audio_bytes)

def _get_voices(self) -> list[dict[str, Any]]:

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 | Confidence: Medium

Hardcoding voice metadata creates a maintenance burden and synchronization risk with the external API. If UpliftAI adds/removes voices, this will require code changes. The related context doesn't show how other engines handle voice metadata, but this static approach is less flexible than dynamic discovery.

Code Suggestion:

def _get_voices(self) -> list[dict[str, Any]]:
        try:
            # Attempt to fetch voices dynamically if API adds endpoint
            response = requests.get(f"{self.BASE_URL}/voices", headers=self.headers)
            return response.json()
        except Exception:
            logger.warning("Falling back to hardcoded voice list")
            return [ ... ]  # Hardcoded fallback

"""Return the list of available voices.

The UpliftAI service does not provide a voices endpoint,
so the voices are hardcoded.
"""
return [
{
"id": "v_meklc281",
"name": "Info/Education V2",
"gender": "neutral",
"language_codes": ["ur"],
},
{
"id": "v_8eelc901",
"name": "Info/Education",
"gender": "neutral",
"language_codes": ["ur"],
},
{
"id": "v_30s70t3a",
"name": "Nostalgic News",
"gender": "neutral",
"language_codes": ["ur"],
},
{
"id": "v_yypgzenx",
"name": "Dada Jee",
"gender": "male",
"language_codes": ["ur"],
},
{
"id": "v_kwmp7zxt",
"name": "Gen Z",
"gender": "neutral",
"language_codes": ["ur"],
},
{"id": "v_sd0kl3m9", "name": "Female", "gender": "female", "language_codes": ["sd"]},
{
"id": "v_sd6mn4p2",
"name": "Male Calm",
"gender": "male",
"language_codes": ["sd"],
},
{
"id": "v_sd9qr7x5",
"name": "Male News",
"gender": "male",
"language_codes": ["sd"],
},
{
"id": "v_bl0ab8c4",
"name": "Balochi Male",
"gender": "male",
"language_codes": ["bal"],
},
{
"id": "v_bl1de2f7",
"name": "Balochi Female",
"gender": "female",
"language_codes": ["bal"],
},
]

def check_credentials(self) -> bool: # pragma: no cover - network call

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 | Confidence: Medium

The credential check performs a full synthesis request ("ping"), which is expensive and slow compared to a dedicated auth endpoint or lighter-weight request. The related context doesn't show how other engines implement credential checks, but this could become problematic if called frequently (e.g., in test setups).

"""Verify that the API key works by making a small request."""
try:
payload = {
"voiceId": self.voice_id or self.DEFAULT_VOICE,
"text": "ping",
"outputFormat": "WAV_22050_16",
}
response = requests.post(self.BASE_URL, json=payload, headers=self.headers, timeout=10)
response.raise_for_status()
return True
except Exception:
logger.debug("UpliftAI credential check failed", exc_info=True)
return False

def connect(self, event_name: str, callback: Callable) -> None:
"""Connect a callback function to an event."""
super().connect(event_name, callback)
Loading
Loading