We present Low Frequency Array (LOFAR) Low Band observations of the Bootes and 3C 295 fields. Our images made at 34, 46, and 62 MHz reach noise levels of 12, 8, and 5 mJy/beam, making them the deepest images ever obtained in this frequency range. In total, we detect between 300 and 400 sources in each of these images, covering an area of 17-52 deg^2^. From the observations, we derive Euclidean-normalized differential source counts. The 62 MHz source counts agree with previous GMRT 153 MHz and Very Large Array 74 MHz differential source counts, scaling with a spectral index of -0.7. We find that a spectral index scaling of -0.5 is required to match up the LOFAR 34 MHz source counts. This result is also in agreement with source counts from the 38 MHz 8C survey, indicating that the average spectral index of radio sources flattens toward lower frequencies. We also find evidence for spectral flattening using the individual flux measurements of sources between 34 and 1400 MHz and by calculating the spectral index averaged over the source population. To select ultra-steep spectrum ({alpha}<-1.1) radio sources that could be associated with massive high-redshift radio galaxies, we compute spectral indices between 62 MHz, 153 MHz, and 1.4 GHz for sources in the Bootes field. We cross-correlate these radio sources with optical and infrared catalogs and fit the spectral energy distribution to obtain photometric redshifts. We find that most of these ultra-steep spectrum sources are located in the 0.7<~z<~2.5 range.